CN115457292A - Greenhouse lighting adjusting method and system - Google Patents
Greenhouse lighting adjusting method and system Download PDFInfo
- Publication number
- CN115457292A CN115457292A CN202211155346.XA CN202211155346A CN115457292A CN 115457292 A CN115457292 A CN 115457292A CN 202211155346 A CN202211155346 A CN 202211155346A CN 115457292 A CN115457292 A CN 115457292A
- Authority
- CN
- China
- Prior art keywords
- feature
- matrix
- feature vector
- training
- characteristic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01G—HORTICULTURE; CULTIVATION OF VEGETABLES, FLOWERS, RICE, FRUIT, VINES, HOPS OR SEAWEED; FORESTRY; WATERING
- A01G9/00—Cultivation in receptacles, forcing-frames or greenhouses; Edging for beds, lawn or the like
- A01G9/24—Devices or systems for heating, ventilating, regulating temperature, illuminating, or watering, in greenhouses, forcing-frames, or the like
- A01G9/249—Lighting means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- H—ELECTRICITY
- H05—ELECTRIC TECHNIQUES NOT OTHERWISE PROVIDED FOR
- H05B—ELECTRIC HEATING; ELECTRIC LIGHT SOURCES NOT OTHERWISE PROVIDED FOR; CIRCUIT ARRANGEMENTS FOR ELECTRIC LIGHT SOURCES, IN GENERAL
- H05B47/00—Circuit arrangements for operating light sources in general, i.e. where the type of light source is not relevant
- H05B47/10—Controlling the light source
- H05B47/17—Operational modes, e.g. switching from manual to automatic mode or prohibiting specific operations
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
- Y02A40/25—Greenhouse technology, e.g. cooling systems therefor
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biodiversity & Conservation Biology (AREA)
- Environmental Sciences (AREA)
- Cultivation Of Plants (AREA)
Abstract
The application discloses a greenhouse lighting adjusting method and a greenhouse lighting adjusting system, wherein a deep neural network model is used for constructing a quality architecture value of a detected plant based on various amino acid values of a detected plant object, the deep neural network model is also used for constructing environment factor architecture values of a plurality of environment factors, and then the quality architecture value and the environment factor architecture value are used as production target values to intelligently adjust a working mode of greenhouse lighting equipment so as to improve the adaptation degree of environmental conditions and plant growth.
Description
Technical Field
The application relates to the technical field of intelligent agriculture, in particular to a greenhouse lighting adjusting method and a greenhouse lighting adjusting system.
Background
In recent years, greenhouse cultivation and smart agriculture are the development direction of new generation crop cultivation. Various devices and devices are arranged in the greenhouse, such as an intelligent control system shown in figure 1, and the intelligent control system comprises a wet curtain fan and an air outlet for adjusting the humidity and the temperature of air; irrigation water equipment for adjusting soil environment factors, a heating belt, N, P and K fertilizer tanks; an air fertilizer tank for conditioning the composition of the air (mainly carbon dioxide in the air); a hot blast stove for raising the temperature of the air; a sunshade net for reducing the light intensity; a fill-in light for prolonging the illumination time or for supplementing the illumination of a certain band, etc. These devices can be turned on and off intelligently to provide better growing conditions for plant production.
However, how to intelligently control the above-mentioned devices to provide better growth conditions for plant growth is a technical problem that needs to be solved urgently, and the difficulty is that different growth conditions are required for plant growth at different stages. One of the existing solutions measures the contents of various amino acids in crops, and constructs a quality calibration value based on the contents of the various amino acids as a representation of the growth state of the plant, so that an environmental target value can be determined based on the quality calibration value, and the equipment can be intelligently controlled based on the environmental target value. However, the effect of the above scheme in actual operation is unsatisfactory, i.e., the plant growth does not meet the predetermined requirement.
The reason is as follows: on the one hand, the substance components of a certain quality of agricultural products, such as the fragrant substances of the muskmelon, can be measured at present for more than 240, but still the quality of the products cannot be accurately described, and the quality cannot be digitally and quantitatively described, because the characteristic of the fragrance is the comprehensive result of the contents of the various components in different proportions. Expensive test equipment, complex test technology and timeliness of samples limit wide development of test work, so that the production practical significance of the quality evaluation method is small.
On the other hand, in the case of quality evaluation based on a plurality of amino acids, the weight of each amino acid component is a preset value, which is not in agreement with the life state of each stage of plant growth. In addition, when the architecture value of the environmental factor is constructed, on one hand, there is a complex linear or non-linear relationship between the environmental factors, for example, temperature affects humidity, and temperature also affects photosynthesis of the plant (that is, temperature also affects the effect of light irradiation).
Therefore, an optimized greenhouse lighting regulation control scheme is expected, which can more accurately express the environmental factors and the quality of agricultural products, thereby improving the adaptability of greenhouse lighting regulation and better promoting the plant growth.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a greenhouse lighting adjusting method and a greenhouse lighting adjusting system, wherein a deep neural network model is used for constructing a quality architecture value of a detected plant based on various amino acid values of the detected plant object, the deep neural network model is also used for constructing environment factor architecture values of a plurality of environment factors, and then the quality architecture value and the environment factor architecture value are used as production target values to intelligently adjust the working mode of greenhouse lighting equipment so as to improve the adaptation degree of environmental conditions and plant growth.
According to one aspect of the application, a greenhouse lighting adjusting method is provided, which comprises the following steps:
obtaining the content values of a plurality of amino acids of the detected plant object at a plurality of preset time points in a preset time period and the measurement values of a plurality of environmental factors at the plurality of preset time points;
passing the content values of the plurality of amino acids of the detected plant object at each preset time point through a trained structural characteristic value encoder with a plurality of fully-connected layers to obtain a structural characteristic value corresponding to each preset time point;
arranging the structural characteristic values corresponding to each preset time point into one-dimensional characteristic vectors according to the time dimension, and then obtaining multi-scale structural characteristic vectors through a trained multi-scale neighborhood characteristic extraction module;
arranging the measured values of the plurality of environmental factors at the plurality of preset time points into an input matrix according to the time dimension and the sample dimension, and then obtaining an environmental factor correlation characteristic matrix by using a convolutional neural network of mutually transposed convolution kernels through adjacent trained layers;
fusing the multi-scale structure feature vector and the environment factor correlation feature matrix to obtain a classification feature vector; and
and enabling the classification characteristic vectors to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether a light supplement lamp in the greenhouse is turned on or not.
According to another aspect of the present application, there is provided a greenhouse lighting adjustment system, comprising:
the data acquisition module is used for acquiring the content values of the plurality of amino acids of the detected plant object at a plurality of preset time points in a preset time period and the measured values of a plurality of environmental factors at the preset time points;
the amino acid content characteristic extraction module is used for enabling the content values of the various amino acids of the detected plant object at each preset time point to pass through a trained structural characteristic value encoder with a plurality of full connection layers so as to obtain a structural characteristic value corresponding to each preset time point;
the multi-scale structure feature vector extraction module is used for arranging the structure feature values corresponding to the preset time points into one-dimensional feature vectors according to the time dimension and then obtaining the multi-scale structure feature vectors through the trained multi-scale neighborhood feature extraction module;
the environment factor feature extraction module is used for arranging the measured values of the plurality of environment factors at the plurality of preset time points into an input matrix according to time dimension and sample dimension, and then obtaining an environment factor correlation feature matrix by using a convolutional neural network of mutually transposed convolution kernels through trained adjacent layers;
the fusion module is used for fusing the multi-scale structure feature vector and the environment factor correlation feature matrix to obtain a classification feature vector; and
and the classification result generation module is used for enabling the classification characteristic vectors to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether a light supplement lamp in the greenhouse is turned on or not.
Compared with the prior art, the greenhouse lighting adjusting method and the greenhouse lighting adjusting system provided by the application use the deep neural network model to construct the quality architecture value of the detected plant based on the multiple amino acid values of the detected plant object, and also use the deep neural network model to construct the environmental factor architecture values of multiple environmental factors, so that the quality architecture value and the environmental factor architecture value are used as the production target values to intelligently adjust the working mode of greenhouse lighting equipment, and the adaptation degree of environmental conditions and plant growth is improved.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is a schematic diagram of a smart system according to an embodiment of the present application.
Fig. 2 is a scene schematic diagram of a greenhouse lighting adjustment method according to an embodiment of the application.
Fig. 3 is a flowchart of a greenhouse lighting adjustment method according to an embodiment of the application.
Fig. 4 is a schematic structural diagram of a greenhouse lighting adjustment method according to an embodiment of the application.
Fig. 5 is a flowchart illustrating the substep of step S120 in the method for adjusting lighting of a greenhouse according to an embodiment of the present disclosure.
Fig. 6 is a flowchart illustrating the sub-step of step S130 in the greenhouse lighting adjustment method according to the embodiment of the application.
Fig. 7 is a flowchart illustrating the sub-step of step S140 in the greenhouse lighting adjustment method according to the embodiment of the application.
Fig. 8 is a flowchart of a substep of further training the structural feature value encoder having multiple fully connected layers, the multi-scale neighborhood feature extraction module, and the adjacent layer using convolutional neural networks of mutually transposed convolutional kernels in the method for adjusting lighting of a greenhouse according to the embodiment of the present application.
Fig. 9 is a block diagram of a greenhouse lighting adjustment system according to an embodiment of the application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Overview of a scene
Accordingly, in the technical solution of the present application, the structure of the crop determines its function. The amino acid contains N-containing substance-amino group and C-containing substance-carboxyl group, and is the basic substance of life activity, and C/N is also the representation of life activity of organism. The system measures the content of amino acids which are basic growth substances, and calculates the framework value by adopting a method of a relational structure, wherein the framework value is equivalent to the pulse condition of the traditional Chinese medicine. The quality of the agricultural product can be calibrated by encoding the architecture value in a current mode, and the digital quantitative expression of the quality is realized. And similarly, obtaining the architecture value of the environmental factor. Therefore, the quality framework value and the environmental factor framework value are used as target values of production, so that the equipment is controlled to operate, a good growth environment is provided, and high-quality agricultural products are produced. The control factors are dynamic, the stability of the architecture value is also dynamic balance, and the instruction sent by the intelligent control system is also dynamic.
Specifically, in the technical solution of the present application, the content values of the plurality of amino acids of the plant object to be tested at a plurality of predetermined time points and the measured values of the environmental factors at the plurality of predetermined time points are obtained first. In a specific example of the application, 18 kinds of amino acids of the plant object to be detected are obtained, and the environmental factors comprise air humidity, air temperature, soil humidity, soil temperature, illumination intensity, quick-acting N, quick-acting P, quick-acting K and CO 2 And (4) concentration.
As described above, in the framework value calculation, if a predetermined weight value is simply given to each amino acid component, this is not in agreement with the life state of each stage of plant production. Therefore, in the technical solution of the present application, the structural feature value encoder having a plurality of fully-connected layers is used to fully-connect encode the content values of the plurality of amino acids of the plant object to be tested at each predetermined time point to obtain the structural feature value corresponding to each predetermined time point. Compared with a statistical model, the structural characteristic value encoder with the multiple fully-connected layers can fully-connected encode the content values of the multiple amino acids of the plant object to be detected at each preset time point so as to fully utilize the content information of each amino acid in the multiple amino acids of the plant object to be detected at each preset time point to improve the accuracy of quantitative evaluation of the structural value.
And then, arranging the structural characteristic values corresponding to the preset time points into one-dimensional characteristic vectors according to the time dimension, and then obtaining the multi-scale structural characteristic vectors through a trained multi-scale neighborhood characteristic extraction module. Here, considering that plant growth is a dynamic organic process, the time series of structural feature values are subjected to one-dimensional convolutional encoding to extract implicit correlation features between multiple structural feature values over different time periods. In particular, considering that plant growth may show different patterns in different time spans, in the technical solution of the present application, a multi-scale neighborhood feature extraction module including a plurality of one-dimensional convolution layers performs one-dimensional convolution coding based on different time scales on the time series of structural feature values to extract implicit correlation features between the structural feature values of plants in different time scales.
As described above, in constructing the architecture value of the environmental factor, on one hand, there is a complex linear or non-linear relationship between the environmental factors, for example, temperature affects humidity, and temperature also affects photosynthesis of the plant (i.e., temperature also affects the effect of light), and therefore, in constructing the architecture value of the environmental factor, if the internal relationship is not taken into consideration, the accuracy of construction of the environmental factor is affected.
Specifically, in the technical solution of the present application, firstly, the measured values of the plurality of environmental factors at the plurality of predetermined time points are arranged as an input matrix according to a time dimension and a sample dimension, that is, the measured values of the plurality of environmental factors at the plurality of predetermined time points are subjected to two-dimensional structuring processing. And then, passing the two-dimensional input matrix through a trained convolutional neural network of which adjacent layers use mutually transposed convolution kernels to obtain an environment factor correlation characteristic matrix. That is, a convolutional neural network model with excellent performance in the local feature extraction field is used as a feature extractor to extract high-dimensional implicit associated features between different environmental factors at the same time point, high-dimensional implicit associated features between different environmental factors at different time points, and high-dimensional implicit features between the same environmental factors at different time points.
In particular, considering that among the plurality of environmental factors, different environmental factors have different degrees of correlation patterns, for example, a strong non-linear correlation exists between temperature and humidity, and the temperature is related to illumination intensity, which requires that the convolutional neural network model has the searching capability of a specific data structure when encoding. Therefore, in the technical solution of the present application, the structure of the convolutional neural network model is adjusted: adjacent layers of the convolutional neural network model use convolutional kernels that are mutually transposed, so that the context factor correlation characteristic matrix can contain correlation characteristics within a predetermined time-sample local structure by using convolutional neural networks of adjacent layers that use convolutional kernels that are mutually transposed.
And then, fusing the multi-scale structural feature vector and the environment factor correlation feature matrix to obtain a classification feature vector containing environment factor architecture value information and plant growth quality architecture value information. And then, the classification result for indicating whether the light supplement lamp is started or not can be obtained by the classification characteristic vector through a classifier, so that the quality framework value and the environmental factor framework value are used as target values of production, equipment is controlled to operate, a good growth environment is provided, and high-quality agricultural products are produced.
In particular, in the technical solution of the present application, by using convolutional neural networks with mutually transposed convolution kernels in adjacent layers, the environment factor associated feature matrix can contain associated features in a predetermined local structure of a time-sample, and since it is necessary to multiply the multi-scale structure feature vector along a time dimension by the environment factor associated feature matrix, it is desirable that the environment factor associated feature matrix has a strong monotonicity in the time dimension, so as to improve a feature fusion degree of the classification feature vector with respect to the multi-scale structure feature vector and the environment factor associated feature matrix.
Therefore, for the environment factor correlation characteristic matrix, a separate parsimonious decomposition incentive loss function is introduced, which is expressed as:
wherein the environmental factor relates to a time dimension of a feature matrixIn the column direction, m i,j And representing the characteristic value of each position of the environment factor associated characteristic matrix, and tau is an overlapping penalty hyperparameter.
The parsimony decomposition encourages the loss function to be used for dividing the characteristic matrix into a group consisting of row vectors, and applying punishment to the overlapping of elements in the row vectors so as to calculate the distance type union of the symbolization function, and promote the geometric shape of the high-dimensional manifold of the environment factor correlation characteristic matrix to be decomposed into a set of convex polyhedrons along the column direction in a parsimony decomposition mode, so that the dimensional monotonicity of the high-dimensional manifold along the column direction is improved.
Based on this, the application provides a greenhouse daylighting adjusting method, which includes: obtaining the content values of a plurality of amino acids of the detected plant object at a plurality of preset time points in a preset time period and the measurement values of a plurality of environmental factors at the preset time points; passing the content values of the plurality of amino acids of the detected plant object at each preset time point through a trained structural characteristic value encoder with a plurality of fully-connected layers to obtain a structural characteristic value corresponding to each preset time point; arranging the structural feature values corresponding to each preset time point into one-dimensional feature vectors according to time dimensions, and then obtaining multi-scale structural feature vectors through a trained multi-scale neighborhood feature extraction module; after the measured values of the environmental factors at the preset time points are arranged into an input matrix according to the time dimension and the sample dimension, a convolutional neural network of mutually transposed convolutional kernels is used by adjacent layers after training is completed to obtain an environmental factor correlation characteristic matrix; fusing the multi-scale structure feature vector and the environment factor correlation feature matrix to obtain a classification feature vector; and enabling the classification characteristic vectors to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether a light supplement lamp in the greenhouse is turned on or not.
Fig. 2 illustrates an application scenario of the greenhouse lighting adjustment method according to the embodiment of the application. As shown in fig. 2, in this application scenario, the content values of the plurality of amino acids of the plant object under test at a plurality of predetermined time points within a predetermined time period are obtained by an amino acid content detection device (e.g., a as illustrated in fig. 2) and the measured values of the plurality of environmental factors at the plurality of predetermined time points are obtained by an environmental factor detection device (e.g., H as illustrated in fig. 2), respectively, and then the obtained content values of the plurality of amino acids of the plant object under test at the plurality of predetermined time points and the measured values of the plurality of environmental factors at the plurality of predetermined time points are input to a server (e.g., S as illustrated in fig. 2) deployed with a greenhouse lighting adjustment algorithm, wherein the server is capable of generating a classification result indicating whether to turn on a fill light (e.g., L as illustrated in fig. 2) within a greenhouse based on the greenhouse lighting adjustment method.
Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary method
Fig. 3 illustrates a flowchart of a greenhouse lighting adjustment method according to an embodiment of the application. As shown in fig. 3, the greenhouse lighting adjustment method according to the embodiment of the application includes the steps of: s110, obtaining the content values of the plurality of amino acids of the detected plant object at a plurality of preset time points in a preset time period and the measurement values of a plurality of environmental factors at the plurality of preset time points; s120, passing the content values of the various amino acids of the detected plant objects at the preset time points through a trained structural characteristic value encoder with a plurality of full-connected layers to obtain structural characteristic values corresponding to the preset time points; s130, arranging the structural feature values corresponding to the preset time points into one-dimensional feature vectors according to the time dimension, and then obtaining the multi-scale structural feature vectors through a trained multi-scale neighborhood feature extraction module; s140, after the measured values of the environmental factors at the preset time points are arranged into an input matrix according to the time dimension and the sample dimension, a convolutional neural network of mutually transposed convolutional kernels is used by adjacent layers after training is completed to obtain an environmental factor correlation characteristic matrix; s150, fusing the multi-scale structure feature vector and the environment factor association feature matrix to obtain a classification feature vector; and S160, enabling the classified characteristic vectors to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether a light supplement lamp in the greenhouse is turned on or not.
Fig. 4 illustrates an architecture diagram of a greenhouse lighting adjustment method according to an embodiment of the application. As shown in fig. 4, in the network architecture, firstly, the content values of the amino acids of the detected plant object at a plurality of predetermined time points in a predetermined time period and the measured values of the environmental factors at the predetermined time points are obtained; then, the content values of the plurality of amino acids of the detected plant object at each preset time point are processed by a trained structure characteristic value encoder with a plurality of full connection layers to obtain the structure characteristic value corresponding to each preset time point; then, arranging the structural feature values corresponding to each preset time point into one-dimensional feature vectors according to the time dimension, and then obtaining multi-scale structural feature vectors through a trained multi-scale neighborhood feature extraction module; then, after the measured values of the environmental factors at the preset time points are arranged into an input matrix according to the time dimension and the sample dimension, a convolutional neural network of mutually transposed convolutional kernels is used by adjacent layers after training is completed to obtain an environmental factor correlation characteristic matrix; then, fusing the multi-scale structure feature vector and the environment factor correlation feature matrix to obtain a classification feature vector; and finally, the classified characteristic vectors are processed by a classifier to obtain a classification result, and the classification result is used for indicating whether a light supplement lamp in the greenhouse is turned on or not.
More specifically, in step S110, the content values of the plurality of amino acids of the plant object to be tested at a plurality of predetermined time points in a predetermined time period and the measured values of the plurality of environmental factors at the plurality of predetermined time points are obtained. The amino acid contains N-containing substances, namely amino and C-containing substance carboxyl, and is a basic substance of life activity, C/N is also a representation of the life activity of an organism, and in a specific example of the application, 18 amino acids of a plant object to be detected are obtained; the environmental factors comprise air humidity, air temperature, soil humidity, soil temperature, illumination intensity, quick-acting N, quick-acting P, quick-acting K and CO2 concentration.
More specifically, in step S120, the content values of the plurality of amino acids of the plant object to be tested at the respective predetermined time points are passed through a trained structural feature value encoder having a plurality of fully-connected layers to obtain the structural feature values corresponding to the respective predetermined time points. In the technical scheme of the application, a structural characteristic value encoder with a plurality of full-connection layers is used for full-connection encoding of the content values of the plurality of amino acids of the detected plant object at each preset time point to obtain the structural characteristic value corresponding to each preset time point. Compared with a statistical model, the structural characteristic value encoder with the multiple fully-connected layers can fully-connected encode the content values of the multiple amino acids of the plant object to be detected at each preset time point so as to fully utilize the content information of each amino acid in the multiple amino acids of the plant object to be detected at each preset time point to improve the accuracy of quantitative evaluation of the structural value.
Accordingly, in a specific example, as shown in fig. 5, the passing the content values of the plurality of amino acids of the plant object to be tested at the respective predetermined time points through a trained structural feature value encoder having a plurality of fully-connected layers to obtain the structural feature value corresponding to the respective predetermined time point includes: s121, arranging the content values of the multiple amino acids of the detected plant object at each preset time point into content input vectors according to sample dimensions; s122, using the structure eigenvalue encoder with multiple fully-connected layers to perform fully-connected encoding on the content input vector to obtain a content eigenvector according to the following formula:
where X is the content input vector, Y is the content feature vector, W is the weight matrix, B is the offset vector,
represents a matrix multiplication; and S123, calculating a global mean value of the feature values of all the positions of the content feature vector as the feature values corresponding to the respective pre-rangesAnd (4) timing the structural characteristic value of the time point.
More specifically, in step S130, the structural feature values corresponding to the predetermined time points are arranged as a one-dimensional feature vector according to a time dimension, and then a multi-scale structural feature vector is obtained through a trained multi-scale neighborhood feature extraction module. Plant growth is a dynamic organic process, and therefore, the time series of structural feature values are subjected to one-dimensional convolutional encoding to extract implicit correlation features between multiple structural feature values over different time periods. In particular, considering that plant growth may show different patterns in different time spans, in the technical solution of the present application, a multi-scale neighborhood feature extraction module including a plurality of one-dimensional convolution layers performs one-dimensional convolution coding based on different time scales on the time series of structural feature values to extract implicit correlation features between the structural feature values of plants in different time scales.
Accordingly, in a specific example, as shown in fig. 6, the obtaining a multi-scale structural feature vector by a trained multi-scale neighborhood feature extraction module after arranging the structural feature values corresponding to the respective predetermined time points into a one-dimensional feature vector according to the time dimension includes: s131, performing one-dimensional convolutional coding on the one-dimensional feature vector by using a first convolutional layer with a first length of the multi-scale neighborhood feature extraction module to obtain a first scale structure feature vector; s132, using a second convolution layer of the multi-scale neighborhood characteristic extraction module to perform one-dimensional convolution coding on the one-dimensional characteristic vector by using a one-dimensional convolution core with a second length to obtain a second scale structure characteristic vector, wherein the second length is different from the first length; and S133, cascading the first scale structure feature vector and the second scale structure feature vector to obtain the multi-scale structure feature vector.
More specifically, in step S140, after the measured values of the plurality of environmental factors at the plurality of predetermined time points are arranged as an input matrix according to the time dimension and the sample dimension, a convolutional neural network of mutually transposed convolution kernels is used by adjacent trained layers to obtain an environmental factor correlation characteristic matrix. It is understood that there is a complex linear or non-linear relationship between the environmental factors, for example, temperature affects humidity, and temperature also affects photosynthesis of plants (i.e., temperature also affects the effect of light), so when constructing the architecture value of the environmental factor, the above-mentioned internal relationship needs to be taken into account to improve the accuracy of construction of the environmental factor. Specifically, in the technical solution of the present application, firstly, the measured values of the plurality of environmental factors at the plurality of predetermined time points are arranged as an input matrix according to a time dimension and a sample dimension, that is, the measured values of the plurality of environmental factors at the plurality of predetermined time points are subjected to two-dimensional structuring processing. And then, passing the input matrix through a trained convolutional neural network of which adjacent layers use mutually transposed convolution kernels to obtain an environment factor correlation characteristic matrix. That is, a convolutional neural network model having excellent performance in the local feature extraction field is used as a feature extractor to extract high-dimensional implicit associated features between different environmental factors at the same time point, high-dimensional implicit associated features between different environmental factors at different time points, and high-dimensional implicit features between the same environmental factors at different time points.
In particular, considering that among the plurality of environmental factors, different environmental factors have different degrees of correlation patterns, for example, a strong non-linear correlation exists between temperature and humidity, and the temperature is related to illumination intensity, which requires that the convolutional neural network model has the searching capability of a specific data structure when encoding. Therefore, in the technical solution of the present application, the structure of the convolutional neural network model is adjusted: adjacent layers of the convolutional neural network model use convolutional kernels which are transposed with each other, so that the environment factor correlation characteristic matrix can contain correlation characteristics in a predetermined time-sample local structure through the convolutional neural network in which adjacent layers use convolutional kernels which are transposed with each other.
Accordingly, in a specific example, as shown in fig. 7, after the measured values of the multiple environmental factors at the multiple predetermined time points are arranged as an input matrix according to the time dimension and the sample dimension, the obtaining the environmental factor correlation feature matrix by using a convolutional neural network in which mutually transposed convolution kernels are used in adjacent trained layers includes: s141, performing convolution processing based on a first convolution kernel, pooling processing along channel dimensions and nonlinear activation processing on the input matrix by using a first layer of the convolutional neural network to obtain a first activation feature map; and S142, performing convolution processing based on a second convolution kernel, pooling processing along a channel dimension, and nonlinear activation processing on the first activation feature map by using a second layer of the convolutional neural network to obtain a second activation feature map, wherein a numerical matrix of the first convolution kernel and a numerical matrix of the second convolution kernel are transposed; and outputting the final layer of the convolutional neural network as the environment factor correlation characteristic matrix.
More specifically, in step S150, the multi-scale structure feature vector and the environment factor associated feature matrix are fused to obtain a classified feature vector. And fusing the multi-scale structural feature vector and the environment factor correlation feature matrix to obtain a classification feature vector containing environment factor architecture value information and plant growth quality architecture value information.
Accordingly, in a specific example, the fusing the multi-scale structure feature vector and the environment factor associated feature matrix to obtain a classified feature vector includes: and multiplying the multi-scale structure feature vector and the environment factor associated feature matrix, and mapping the high-dimensional feature information of the environment factor associated feature matrix into a high-dimensional feature domain of the multi-scale structure feature vector to obtain the classified feature vector.
Accordingly, in another specific example, the fusing the multi-scale structure feature vector and the environment factor associated feature matrix to obtain a classified feature vector includes: and calculating a transfer vector of the multi-scale structure feature vector relative to the environment factor associated feature matrix as the classification feature vector.
More specifically, in step S160, the classified feature vectors are passed through a classifier to obtain a classification result, where the classification result is used to indicate whether to turn on a fill-in light in the greenhouse. And the classification result for indicating whether the light supplement lamp is started or not can be obtained by the classification characteristic vector through a classifier, so that the quality architecture value and the environmental factor architecture value are used as target values of production, equipment is controlled to operate, a good growth environment is provided, and high-quality agricultural products are produced.
Accordingly, in a specific example, as shown in fig. 8, the greenhouse lighting adjustment method further includes: training the structural feature value encoder with the multiple fully-connected layers, the multi-scale neighborhood feature extraction module and the adjacent layer by using a convolutional neural network of mutually transposed convolution kernels; wherein the process of training the structural feature value encoder having a plurality of fully connected layers, the multi-scale neighborhood feature extraction module, and the adjacent layer using a convolutional neural network in which convolutional kernels that are transposed with each other includes: s210, obtaining training data, wherein the training data comprises content values of a plurality of amino acids of the detected plant object at a plurality of preset time points in a preset time period and measurement values of a plurality of environmental factors at the plurality of preset time points; s220, passing the content values of the plurality of amino acids of the detected plant object at each preset time point through the structure characteristic value encoder with the plurality of full-connection layers to obtain training structure characteristic values corresponding to each preset time point; s230, arranging the training structure characteristic values corresponding to the preset time points into one-dimensional characteristic vectors according to the time dimension, and then obtaining training multi-scale structure characteristic vectors through the multi-scale neighborhood characteristic extraction module; s240, arranging the measured values of the environmental factors at the preset time points into an input matrix according to the time dimension and the sample dimension, and then obtaining a training environmental factor correlation characteristic matrix by using a convolutional neural network of mutually transposed convolution kernels through trained adjacent layers; s250, fusing the training multi-scale structure feature vector and the training environment factor association feature matrix to obtain a training classification feature vector; s260, enabling the training classification feature vector to pass through the classifier to obtain a classification loss function value; s270, calculating a parsimonious decomposition incentive loss function value of the training environment factor correlation characteristic matrix; and S280, training the structural feature value encoder with the multiple fully-connected layers, the multi-scale neighborhood feature extraction module and the adjacent layer by using the weighted sum of the frugal incentive loss function value and the classification loss function value as a loss function value, wherein the convolutional neural network is formed by mutually transposing convolution kernels.
Accordingly, in one specific example, the passing the training classification feature vector through the classifier to obtain a classification loss function value includes: inputting the training classification feature vector into a Softmax classification function of the classifier to obtain a classification result; and calculating a cross entropy value between the classification result and a real value as the classification loss function value.
In the technical solution of the present application, through a convolutional neural network in which adjacent layers use convolution kernels that are transposed to each other, the environment factor associated feature matrix can contain associated features within a predetermined local structure of a time-sample, and since it is necessary to multiply the multi-scale structure feature vector along a time dimension by the environment factor associated feature matrix, it is desirable that the environment factor associated feature matrix can have a strong monotonicity in the time dimension, so as to improve a feature fusion degree of the classification feature vector with respect to the multi-scale structure feature vector and the environment factor associated feature matrix.
Thus, in one specific example, the computing a parsimonious decomposition of the training environment factor correlation feature matrix encourages loss function values, comprising: calculating the frugal decomposition incentive loss function value of the training environmental factor correlation feature matrix with the following formula;
wherein the formula is:
wherein m is i,j Representing an eigenvalue of each position of the training environment factor correlation eigenmatrix, and τ is an overlapPunishment over-parameter, | \ | live in the sky 2 Representing the two-norm of the feature vector. The parsimonious encouragement loss function is used for dividing the feature matrix into groups consisting of row vectors and applying penalties to overlapping of elements within the row vectors to compute a distance-wise union of the symbolization function to promote the parsimonious decomposition of the geometry of the high-dimensional manifold of the environment factor associated feature matrix into a set of convex polyhedrons along the column direction in the form of parsimonious decomposition, thereby improving the dimensional monotonicity of the high-dimensional manifold along the column direction.
In summary, according to the greenhouse lighting adjustment method and system provided by the embodiment of the application, the deep neural network model is used, the quality architecture value of the detected plant is constructed based on the multiple amino acid values of the detected plant object, the deep neural network model is also used to construct the environmental factor architecture values of the multiple environmental factors, and then the quality architecture value and the environmental factor architecture value are used as the production target values to intelligently adjust the working mode of the greenhouse lighting equipment, so as to improve the adaptability of the environmental conditions and the plant growth.
Exemplary System
Fig. 9 illustrates a block diagram of a greenhouse lighting adjustment system 100 according to an embodiment of the present application. As shown in fig. 9, a greenhouse lighting adjustment system 100 according to an embodiment of the present application includes: a data obtaining module 110, configured to obtain content values of a plurality of amino acids of a plant object to be tested at a plurality of predetermined time points and measurement values of a plurality of environmental factors at the plurality of predetermined time points within a predetermined time period; an amino acid content feature extraction module 120, configured to pass the content values of the plurality of amino acids of the detected plant object at each predetermined time point through a trained structural feature value encoder having a plurality of fully-connected layers to obtain a structural feature value corresponding to each predetermined time point; a multi-scale structure feature vector extraction module 130, configured to arrange the structure feature values corresponding to each predetermined time point into one-dimensional feature vectors according to a time dimension, and then obtain multi-scale structure feature vectors through a trained multi-scale neighborhood feature extraction module; the environmental factor feature extraction module 140 is configured to arrange the measured values of the environmental factors at the multiple predetermined time points into an input matrix according to a time dimension and a sample dimension, and then obtain an environmental factor correlation feature matrix by using a convolutional neural network in which mutually transposed convolution kernels are used in adjacent trained layers; a fusion module 150, configured to fuse the multi-scale structure feature vector and the environment factor correlation feature matrix to obtain a classification feature vector; and the classification result generation module 160 is configured to pass the classification feature vectors through a classifier to obtain a classification result, where the classification result is used to indicate whether to turn on a light supplement lamp in the greenhouse.
In an example, in the greenhouse lighting adjustment system 100, the amino acid content feature extraction module 120 is configured to: arranging the content values of the multiple amino acids of the detected plant object at each preset time point into content input vectors according to the dimension of the sample; performing full-concatenation encoding on the content input vector to obtain a content feature vector by using the structure feature value encoder with a plurality of full-concatenation layers according to the following formula:
where X is the content input vector, Y is the content feature vector, W is the weight matrix, B is the offset vector,
represents a matrix multiplication; and calculating a global mean value of the feature values of all positions of the content feature vector as the structural feature value corresponding to each predetermined time point.
In an example, in the above greenhouse lighting adjustment system 100, the multi-scale structural feature vector extraction module 130 is configured to: performing one-dimensional convolution coding on the one-dimensional feature vector by using a first convolution layer of the multi-scale neighborhood feature extraction module and a one-dimensional convolution core with a first length to obtain a first scale structural feature vector; performing one-dimensional convolution encoding on the one-dimensional feature vector by using a second convolution layer of the multi-scale neighborhood feature extraction module by using a one-dimensional convolution kernel with a second length to obtain a second scale structural feature vector, wherein the second length is different from the first length; and cascading the first scale structure feature vector and the second scale structure feature vector to obtain the multi-scale structure feature vector.
In an example, in the greenhouse lighting adjustment system 100, the environmental factor feature extraction module 140 is configured to: performing convolution processing based on a first convolution kernel, pooling processing along channel dimensions, and nonlinear activation processing on the input matrix using a first layer of the convolutional neural network to obtain a first activation feature map; performing convolution processing based on a second convolution kernel, pooling processing along a channel dimension and nonlinear activation processing on the first activation characteristic graph by using a second layer of the convolutional neural network to obtain a second activation characteristic graph, wherein a numerical matrix of the first convolution kernel and a numerical matrix of the second convolution kernel are transposes; wherein the output of the last layer of the convolutional neural network is the environment factor correlation characteristic matrix.
In an example, in the above greenhouse lighting adjustment system 100, the fusion module 150 is configured to: and multiplying the multi-scale structure feature vector by the environment factor correlation feature matrix, and mapping the high-dimensional feature information of the environment factor correlation feature matrix into a high-dimensional feature domain of the multi-scale structure feature vector to obtain the classification feature vector.
In another example, in the above greenhouse lighting adjustment system 100, the fusion module 150 is configured to: and calculating a transfer vector of the multi-scale structure feature vector relative to the environment factor associated feature matrix as the classification feature vector.
In an example, the greenhouse lighting adjustment system 100 further includes a training module for training the structural feature value encoder having a plurality of fully connected layers, the multi-scale neighborhood feature extraction module, and the adjacent layer using a convolutional neural network with mutually transposed convolutional kernels. Training the structural characteristic value encoder with the multiple fully-connected layers, the multi-scale neighborhood characteristic extraction module and the adjacent layer by using the convolutional neural networks of the mutually transposed convolutional kernels to update parameters of the structural characteristic value encoder with the multiple fully-connected layers, the multi-scale neighborhood characteristic extraction module and the adjacent layer by using the convolutional neural networks of the mutually transposed convolutional kernels, and finally enabling the greenhouse lighting adjustment system 100 to judge whether to turn on a light supplement lamp in the greenhouse more accurately, so that a better growth environment is provided, and agricultural products with higher quality are produced.
More specifically, in an embodiment of the present application, the training module includes: a training data acquisition unit, configured to acquire training data, where the training data includes content values of a plurality of amino acids of the plant object to be tested at a plurality of predetermined time points within a predetermined time period and measurement values of a plurality of environmental factors at the plurality of predetermined time points; the training structure characteristic value extraction unit is used for enabling the content values of the multiple amino acids of the detected plant object at each preset time point to pass through the structure characteristic value encoder with the multiple full-connection layers to obtain a training structure characteristic value corresponding to each preset time point; the training multi-scale structure feature vector generating unit is used for arranging the training structure feature values corresponding to the preset time points into one-dimensional feature vectors according to time dimensions and then obtaining training multi-scale structure feature vectors through the multi-scale neighborhood feature extraction module; the training environment factor correlation characteristic matrix generating unit is used for arranging the measured values of the plurality of environment factors at the plurality of preset time points into an input matrix according to the time dimension and the sample dimension, and then obtaining a training environment factor correlation characteristic matrix by using a convolutional neural network of mutually transposed convolutional kernels through adjacent layers after training; the training classification feature vector generating unit is used for fusing the training multi-scale structure feature vector and the training environment factor association feature matrix to obtain a training classification feature vector; the classification loss function value generating unit is used for enabling the training classification characteristic vector to pass through the classifier so as to obtain a classification loss function value; the frugal decomposition incentive loss function value calculation unit is used for calculating a frugal decomposition incentive loss function value of the training environment factor correlation characteristic matrix; and a training unit, configured to train the structural feature value encoder with multiple fully-connected layers, the multi-scale neighborhood feature extraction module, and the adjacent layer using a convolutional neural network with mutually transposed convolution kernels, with a weighted sum of the parsimonious decomposition incentive loss function value and the classification loss function value as a loss function value.
More specifically, in an embodiment of the present application, the classification loss function value generating unit is configured to: inputting the training classification feature vector into a Softmax classification function of the classifier to obtain a classification result; and calculating a cross entropy value between the classification result and a real value as the classification loss function value.
More specifically, in the embodiment of the present application, the parsimonious incentive loss function value calculation unit calculates the parsimonious incentive loss function value of the training environment factor associated feature matrix with the following formula; wherein the formula is:
wherein m is i,j Representing the characteristic value of each position of the training environment factor correlation characteristic matrix, and tau is an overlapping punishment hyperparameter, | · | sweet 2 Representing the two-norm of the feature vector.
Here, it can be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the greenhouse lighting adjustment system 100 described above have been described in detail in the description of the greenhouse lighting adjustment method with reference to fig. 2 to 8, and thus, a repetitive description thereof will be omitted.
As described above, the greenhouse lighting adjustment system 100 according to the embodiment of the present application may be implemented in various wireless terminals, such as a server of the greenhouse lighting adjustment system. In one example, the greenhouse lighting adjustment system 100 according to the embodiment of the present application can be integrated into a wireless terminal as a software module and/or a hardware module. For example, the greenhouse lighting adjustment system 100 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the greenhouse lighting adjustment system 100 can also be one of many hardware modules of the wireless terminal.
Alternatively, in another example, the greenhouse lighting adjustment system 100 and the wireless terminal may be separate devices, and the greenhouse lighting adjustment system 100 may be connected to the wireless terminal through a wired and/or wireless network and transmit the interaction information according to an agreed data format.
The basic principles of the present application have been described above with reference to specific embodiments, but it should be noted that advantages, effects, etc. mentioned in the present application are only examples and are not limiting, and the advantages, effects, etc. must not be considered to be possessed by various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.
The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. As used herein, the words "or" and "refer to, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".
It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations are to be considered as equivalents of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.
Claims (10)
1. A greenhouse lighting adjusting method is characterized by comprising the following steps:
obtaining the content values of a plurality of amino acids of the detected plant object at a plurality of preset time points in a preset time period and the measurement values of a plurality of environmental factors at the plurality of preset time points;
passing the content values of the plurality of amino acids of the detected plant object at each preset time point through a trained structural characteristic value encoder with a plurality of fully-connected layers to obtain a structural characteristic value corresponding to each preset time point;
arranging the structural feature values corresponding to each preset time point into one-dimensional feature vectors according to time dimensions, and then obtaining multi-scale structural feature vectors through a trained multi-scale neighborhood feature extraction module;
arranging the measured values of the plurality of environmental factors at the plurality of preset time points into an input matrix according to the time dimension and the sample dimension, and then obtaining an environmental factor correlation characteristic matrix by using a convolutional neural network of mutually transposed convolution kernels through adjacent trained layers;
fusing the multi-scale structure feature vector and the environment factor correlation feature matrix to obtain a classification feature vector; and
and passing the classified characteristic vectors through a classifier to obtain a classification result, wherein the classification result is used for indicating whether a light supplement lamp in the greenhouse is turned on or not.
2. The greenhouse lighting regulation method of claim 1, wherein the step of passing the content values of the plurality of amino acids of the plant subject to be tested at the predetermined time points through a trained structural feature value encoder having a plurality of fully-connected layers to obtain the structural feature value corresponding to each of the predetermined time points comprises:
arranging the content values of the multiple amino acids of the detected plant object at each preset time point into content input vectors according to the dimension of the sample;
using the structure eigenvalue encoder with a plurality of fully-connected layers to fully-connect encode the content input vector to obtain a content eigenvector according to the following formula:
where X is the content input vector, Y is the content feature vector, W is the weight matrix, B is the offset vector,
represents a matrix multiplication; and
and calculating a global mean value of the characteristic values of all the positions of the content characteristic vector as the structural characteristic value corresponding to each preset time point.
3. The lighting adjustment method for greenhouses according to claim 2, wherein the step of arranging the structural feature values corresponding to each predetermined time point into one-dimensional feature vectors according to time dimension and then obtaining the multi-dimensional structural feature vectors through a trained multi-scale neighborhood feature extraction module comprises:
performing one-dimensional convolution coding on the one-dimensional feature vector by using a first convolution layer of the multi-scale neighborhood feature extraction module through a one-dimensional convolution core with a first length to obtain a first scale structure feature vector;
performing one-dimensional convolutional encoding on the one-dimensional feature vector by using a second convolutional layer of the multi-scale neighborhood feature extraction module by using a one-dimensional convolutional layer with a second length to obtain a second scale structure feature vector, wherein the second length is different from the first length; and
and cascading the first scale structure feature vector and the second scale structure feature vector to obtain the multi-scale structure feature vector.
4. The greenhouse lighting adjustment method according to claim 3, wherein the step of arranging the measured values of the plurality of environmental factors at the plurality of predetermined time points into an input matrix according to the time dimension and the sample dimension and then obtaining the environmental factor associated feature matrix by using a trained convolutional neural network with mutually transposed convolution kernels for adjacent layers comprises the steps of:
performing convolution processing based on a first convolution kernel, pooling processing along channel dimensions, and nonlinear activation processing on the input matrix using a first layer of the convolutional neural network to obtain a first activation profile; and
performing convolution processing based on a second convolution kernel, pooling processing along a channel dimension and nonlinear activation processing on the first activation characteristic graph by using a second layer of the convolutional neural network to obtain a second activation characteristic graph, wherein a numerical matrix of the first convolution kernel and a numerical matrix of the second convolution kernel are transposes;
wherein the output of the last layer of the convolutional neural network is the environment factor correlation characteristic matrix.
5. The greenhouse lighting adjustment method according to claim 4, wherein the fusing the multi-scale structural feature vector and the environment factor association feature matrix to obtain a classification feature vector comprises:
and multiplying the multi-scale structure feature vector by the environment factor correlation feature matrix, and mapping the high-dimensional feature information of the environment factor correlation feature matrix into a high-dimensional feature domain of the multi-scale structure feature vector to obtain the classification feature vector.
6. The lighting adjustment method for greenhouses according to claim 4, wherein the fusing the multi-scale structural feature vector and the environment factor correlation feature matrix to obtain a classification feature vector comprises:
and calculating a transfer vector of the multi-scale structure feature vector relative to the environment factor associated feature matrix as the classification feature vector.
7. The greenhouse lighting adjustment method according to claim 1, further comprising: training the structural feature value encoder with the multiple fully connected layers, the multi-scale neighborhood feature extraction module and the adjacent layer by using a convolutional neural network of mutually transposed convolution kernels;
wherein the process of training the structural feature value encoder having a plurality of fully connected layers, the multi-scale neighborhood feature extraction module, and the adjacent layer using a convolutional neural network in which convolutional kernels that are transposed with each other includes:
obtaining training data, wherein the training data comprises content values of a plurality of amino acids of a detected plant object at a plurality of preset time points in a preset time period and measurement values of a plurality of environmental factors at the preset time points;
passing the content values of the plurality of amino acids of the detected plant object at each predetermined time point through the structural characteristic value encoder with a plurality of fully-connected layers to obtain a training structural characteristic value corresponding to each predetermined time point;
arranging the training structure characteristic values corresponding to each preset time point into one-dimensional characteristic vectors according to the time dimension, and then obtaining training multi-scale structure characteristic vectors through the multi-scale neighborhood characteristic extraction module;
arranging the measured values of the environmental factors of the preset time points into an input matrix according to the time dimension and the sample dimension, and then obtaining a training environmental factor correlation characteristic matrix by using a convolutional neural network of mutually transposed convolution kernels through adjacent layers after training;
fusing the training multi-scale structure feature vector and the training environment factor correlation feature matrix to obtain a training classification feature vector;
passing the training classification feature vector through the classifier to obtain a classification loss function value;
calculating a frugal decomposition incentive loss function value of the training environment factor correlation characteristic matrix; and
training the structural feature value encoder with the plurality of fully-connected layers, the multi-scale neighborhood feature extraction module, and the adjacent layer using convolution neural networks with convolution kernels that are transposed with respect to each other with a weighted sum of the parsimonious decomposition incentive loss function values and the classification loss function values as loss function values.
8. The greenhouse lighting adjustment method according to claim 7, wherein the passing the training classification feature vector through the classifier to obtain a classification loss function value comprises:
inputting the training classification feature vector into a Softmax classification function of the classifier to obtain a classification result; and
and calculating a cross entropy value between the classification result and the real value as the classification loss function value.
9. The greenhouse lighting adjustment method according to claim 8, wherein the calculating a parsimonious decomposition incentive loss function value of the training environment factor correlation feature matrix comprises:
calculating the frugal factorization incentive loss function value of the training environmental factor correlation feature matrix with the following formula;
wherein the formula is:
wherein m is i,j Representing the characteristic value of each position of the training environment factor correlation characteristic matrix, and tau is an overlapping punishment hyperparameter, | · | sweet 2 Representing the two-norm of the feature vector.
10. A greenhouse daylighting governing system which characterized in that includes:
the data acquisition module is used for acquiring the content values of the plurality of amino acids of the detected plant object at a plurality of preset time points in a preset time period and the measured values of a plurality of environmental factors at the preset time points;
the amino acid content characteristic extraction module is used for enabling the content values of the multiple amino acids of the detected plant object at each preset time point to pass through a trained structural characteristic value encoder with multiple full-connection layers to obtain a structural characteristic value corresponding to each preset time point;
the multi-scale structural feature vector extraction module is used for arranging the structural feature values corresponding to the preset time points into one-dimensional feature vectors according to time dimensions and then obtaining the multi-scale structural feature vectors through the trained multi-scale neighborhood feature extraction module;
the environment factor characteristic extraction module is used for arranging the measured values of the plurality of environment factors at the plurality of preset time points into an input matrix according to time dimension and sample dimension, and then obtaining an environment factor correlation characteristic matrix by using a convolutional neural network of mutually transposed convolution kernels through adjacent layers after training is finished;
the fusion module is used for fusing the multi-scale structure feature vector and the environment factor association feature matrix to obtain a classification feature vector; and
and the classification result generation module is used for enabling the classification characteristic vectors to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether a light supplement lamp in the greenhouse is turned on or not.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211155346.XA CN115457292A (en) | 2022-09-22 | 2022-09-22 | Greenhouse lighting adjusting method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211155346.XA CN115457292A (en) | 2022-09-22 | 2022-09-22 | Greenhouse lighting adjusting method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115457292A true CN115457292A (en) | 2022-12-09 |
Family
ID=84305912
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211155346.XA Pending CN115457292A (en) | 2022-09-22 | 2022-09-22 | Greenhouse lighting adjusting method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115457292A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116759053A (en) * | 2023-06-16 | 2023-09-15 | 烟台汇通佳仁医疗科技有限公司 | Medical system prevention and control method and system based on Internet of things system |
| CN118135407A (en) * | 2024-05-07 | 2024-06-04 | 中邦生态环境有限公司 | Arbor nutrient solution preparation dynamic adjustment system and method based on big data |
-
2022
- 2022-09-22 CN CN202211155346.XA patent/CN115457292A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116759053A (en) * | 2023-06-16 | 2023-09-15 | 烟台汇通佳仁医疗科技有限公司 | Medical system prevention and control method and system based on Internet of things system |
| CN118135407A (en) * | 2024-05-07 | 2024-06-04 | 中邦生态环境有限公司 | Arbor nutrient solution preparation dynamic adjustment system and method based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115457292A (en) | Greenhouse lighting adjusting method and system | |
| Chia et al. | Recent advances in evapotranspiration estimation using artificial intelligence approaches with a focus on hybridization techniques—a review | |
| Zhuo et al. | Consumptive water footprint and virtual water trade scenarios for China—With a focus on crop production, consumption and trade | |
| Li et al. | Estimating crop yield from multi-temporal satellite data using multivariate regression and neural network techniques | |
| GB2596886A (en) | Method and device for performing inversion of crop leaf area index | |
| Gong et al. | Multi-objective parameter optimization of common land model using adaptive surrogate modeling | |
| CN112906300B (en) | Polarization SAR soil humidity inversion method based on double-channel convolutional neural network | |
| CN110579186B (en) | Crop growth monitoring method based on inversion of leaf area index by inverse Gaussian process | |
| CN110069032A (en) | A kind of eggplant greenhouse intelligent checking system based on wavelet neural network | |
| CN111523525A (en) | Crop classification identification method and device and electronic equipment | |
| CN117073768B (en) | Beef cattle cultivation management system and method thereof | |
| CN113435707A (en) | Soil testing and formulated fertilization method based on deep learning and weighted multi-factor evaluation | |
| CN110119767A (en) | A kind of cucumber green house temperature intelligent detection device based on LVQ neural network | |
| Fei et al. | Bayesian model averaging to improve the yield prediction in wheat breeding trials | |
| CN112084991A (en) | Crop early identification method based on multi-source remote sensing time sequence image and convolutional neural network | |
| Jaihuni et al. | Determining spatiotemporal distribution of macronutrients in a cornfield using remote sensing and a deep learning model | |
| Çetiner et al. | Recurrent neural network based model development for wheat yield forecasting | |
| von Bloh et al. | Machine learning for soybean yield forecasting in Brazil | |
| CN113408700A (en) | Convolutional neural network soil alkaline hydrolysis nitrogen analysis model construction system and method | |
| Jin et al. | Research on the rice fertiliser decision-making method based on UAV remote sensing data assimilation | |
| CN117541422A (en) | Agricultural production management system based on agricultural environment data | |
| CN116629453B (en) | Remote sensing yield estimation method suitable for whole crop growth period | |
| CN110782112B (en) | Method and system for estimating greenhouse gas emission reduction potential in crop production | |
| Monica et al. | Soil NPK prediction using enhanced genetic algorithm | |
| CN109472320B (en) | Method for automatically correcting variety parameters of crop growth period model under uncertain conditions |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |