CN112464766A - Farmland automatic identification method and system - Google Patents
Farmland automatic identification method and system Download PDFInfo
- Publication number
- CN112464766A CN112464766A CN202011287841.7A CN202011287841A CN112464766A CN 112464766 A CN112464766 A CN 112464766A CN 202011287841 A CN202011287841 A CN 202011287841A CN 112464766 A CN112464766 A CN 112464766A
- Authority
- CN
- China
- Prior art keywords
- farmland
- image
- network model
- ground
- headland
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 75
- 238000012549 training Methods 0.000 claims abstract description 53
- 239000002689 soil Substances 0.000 claims abstract description 14
- 238000003860 storage Methods 0.000 claims description 15
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 13
- 238000005520 cutting process Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 5
- 230000008447 perception Effects 0.000 abstract description 4
- 230000008569 process Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 9
- 238000012795 verification Methods 0.000 description 7
- 239000011159 matrix material Substances 0.000 description 5
- 238000013526 transfer learning Methods 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000002265 prevention Effects 0.000 description 3
- 241000196324 Embryophyta Species 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 235000008733 Citrus aurantifolia Nutrition 0.000 description 1
- 241001118108 Isatis quadrialata Species 0.000 description 1
- 235000000903 Ranunculus bulbosus Nutrition 0.000 description 1
- 235000011941 Tilia x europaea Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000004571 lime Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035939 shock Effects 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000007306 turnover Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the invention provides a farmland heading automatic identification method and a farmland heading automatic identification system, wherein the farmland heading automatic identification method comprises the following steps: obtaining a farmland head image; inputting the farmland head image into the head recognition network model, and obtaining a farmland operation environment category corresponding to the farmland head image according to an output result of the head recognition network model; the farmland heading network model is obtained by training farmland heading image samples with farmland operation environment class labels; the farmland operation environment categories comprise non-crop farmland, green vegetation land, yellow vegetation land, bare soil land and artificial facility land. The farmland ground automatic identification method and the farmland ground automatic identification system provided by the embodiment of the invention are used for identifying the farmland ground image under the agricultural machinery operation scene based on deep learning, can accurately acquire the operation environment of the current agricultural machinery operation, are good in real-time performance, and can provide a feasible technical solution for autonomous perception of the farmland environment of the intelligent agricultural machinery.
Description
Technical Field
The invention relates to the field of agricultural machinery, in particular to a farmland heading automatic identification method and device.
Background
The field operation process of the agricultural machine can be understood as follows: the agricultural machine is directed to the regular movement of a specific target within a specific range. Although the scale of the farmland plot and the form of the agricultural machine set are different, each agricultural machine set necessarily needs to perform the turning operation of the farmland, so the turning region of the farmland is the boundary of the operation range of the agricultural machine. The agricultural machine automatic navigation technology can obviously reduce the overlapping and omission between reciprocating operation ridges, reduces the labor intensity of drivers, and the current practical agricultural machine automatic navigation system is still in a semi-automatic stage and still needs manual operation at the turning of the ground.
One way to implement automatic navigation of agricultural machinery in the prior art is based on a digital map of the farmland. The farmland plots are drawn by surveying and mapping or remote sensing image modes, and the plot boundaries are abstracted into linear attributes in the digital map of the farmland plots. However, in fact, the ground for turning agricultural machinery is an area, so that the precision of the current farmland digital map cannot meet the requirement of automatic driving of agricultural machinery.
Another approach in the prior art for implementing automatic navigation of agricultural machinery is based on image recognition technology. The key step of image recognition is image feature extraction, and the traditional image feature extraction mode is to use an artificially designed extractor to extract local features such as color, texture, shape and the like through professional knowledge and a complex parameter adjusting process. In addition, the actual farmland headers are complex and various in types, and may be the boundaries of adjacent farmland plots for planting crops or non-farmland areas (such as ridges, lime roads, ditches or bare soil and the like), so that the artificial feature extraction mode adopting the traditional image recognition has many limitations and cannot well adapt to the characteristics of complex and variable farmland header environments under natural conditions.
Disclosure of Invention
In order to overcome the defect that the farmland environment under natural conditions cannot be well adapted to be complex and changeable in the prior art, the embodiment of the invention provides an automatic farmland environment identification method and device, which can acquire and automatically identify farmland images under an agricultural machinery operation scene so as to meet the application requirement of intelligent agricultural machinery environment sensing.
In a first aspect, an embodiment of the present invention provides a method for automatically identifying a farmland, which mainly includes: obtaining a farmland head image; inputting the farmland head image into the head recognition network model, and obtaining a farmland operation environment category corresponding to the farmland head image according to an output result of the head recognition network model; the farmland heading network model is obtained by training farmland heading image samples with farmland operation environment class labels; the farmland operation environment categories comprise non-crop farmland, green vegetation land, yellow vegetation land, bare soil land and artificial facility land.
Optionally, after acquiring the farmland head image, the method further includes: and cutting the farmland head image into an aspect ratio of 1 by taking the short side of the farmland head image as the side length: 1; zooming the first image based on a nearest neighbor interpolation method to obtain a second image; the pixels of the second image are identifiable pixels of the geography-identifying network model.
Optionally, the head-of-ground recognition network model is based on MobileNet V2And (4) constructing.
Optionally, the network structure of the head recognition network model includes an input layer, a first convolution module, an invoked redundant Block module, a second convolution module, a global averaging pooling layer, a Dropout layer, a 1 × 1convolution layer, and a Softmax layer, which are connected in sequence; the first convolution module and the second convolution module have the same structure and respectively comprise a convolution layer, a batch normalization layer and a Relu activation layer which are sequentially connected; the invoked Residual Block module comprises 7 invoked Residual blocks which are connected in sequence; shortcuts are used between convolutional layers with a sliding step of 1 in each invoked Residual Block.
Optionally, before inputting the farmland head image into the head recognition network model, the method further comprises: obtaining a plurality of farmland ground image samples of each farmland operation environment type; cutting the farmland headland image samples according to a preset size proportion by taking the image top of each farmland headland image sample as a cutting starting point in a centrosymmetric mode to obtain multi-size image samples; turning each farmland ground image sample according to a preset angle to obtain multi-angle image samples; randomly transforming the brightness, contrast and chromaticity of each farmland head image sample to obtain a multicolor image sample; constructing a farmland head image sample set consisting of the farmland head image sample, the multi-size image sample, the multi-angle image sample and the multi-color image sample; each image sample in the farmland head image sample set corresponds to a farmland operation environment type label; and pre-training the ground recognition network model by utilizing the farmland ground image sample set.
Optionally, after obtaining a plurality of farmland image samples of each farmland operation environment category, the method further includes preprocessing each image sample in the farmland image sample set, including: and performing cutting processing and scaling processing so that the processed pixel of each image sample is an identifiable pixel of the ground identification network model.
Optionally, before the pre-training of the farmland headland image sample set on the headland recognition network model, the method further includes: and pre-training the ground recognition network model by taking ImageNet as a source data set.
In a second aspect, an embodiment of the present invention further provides an automatic farmland heading identification device, which mainly includes a sensor module and a microprocessor module, wherein:
the sensor module is mainly used for acquiring a farmland head image; the microprocessor module is loaded with a ground identification network model which is mainly used for receiving farmland ground images so as to output farmland operation environment categories corresponding to the farmland ground images.
The farmland heading identification network model is obtained by training farmland heading image samples with farmland operation environment class labels; the farmland operation environment categories comprise non-crop farmlands, green vegetation lands, yellow vegetation lands, bare soil lands and artificial facilities lands.
In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of any one of the above-mentioned methods for automatically identifying a farmland heading when executing the program.
In a fourth aspect, an embodiment of the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the automatic farmland heading identification method according to any one of the above-mentioned methods.
The farmland headland automatic identification method and device provided by the embodiment of the invention are used for identifying the farmland headland image in the agricultural machinery operation scene based on deep learning, can accurately acquire the operation environment of the current agricultural machinery operation, are good in real-time performance, and can provide a feasible technical solution for autonomous perception of the farmland environment of the intelligent agricultural machinery.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of a method for automatically identifying farmland headers provided by an embodiment of the invention;
FIG. 2 is a schematic diagram of classification of farmland environment provided by an embodiment of the invention;
fig. 3 is a schematic diagram of a network structure of MobileNetV2 according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of an automatic farmland heading identification device provided by an embodiment of the invention;
FIG. 5 is a schematic flow chart of another farmland over-the-ground automatic identification method provided by the embodiment of the invention;
FIG. 6 is a visual confusion matrix diagram of the farmland automatic identification result provided by the embodiment of the invention;
FIG. 7 is a schematic structural diagram of another automatic farmland heading identification device provided by an embodiment of the invention;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of an automatic farmland heading identification method provided by an embodiment of the present invention, as shown in fig. 1, including but not limited to the following steps:
step S1: obtaining a farmland head image;
step S2: and inputting the farmland head image into a head recognition network model, and obtaining the farmland operation environment category corresponding to the farmland head image according to the output result of the head recognition network model.
The farmland heading identification network model is obtained after training according to farmland heading image samples with farmland operation environment class labels.
The farmland operation environment categories comprise non-crop farmlands, green vegetation lands, yellow vegetation lands, bare soil lands and artificial facilities lands.
Wherein, the farmland ground image can be shot in real time by a high-resolution camera arranged on the working agricultural machine.
In order to better implement the application of automatic identification of the farmland environment, the camera is preferably designed to be a three-prevention industrial-grade firm shell with dust prevention, shock prevention and water prevention, so that the use requirement of the vehicle-mounted terminal of the agricultural machinery is met. In addition, the camera is suggested to be fixed above the agricultural machine cab and installed at a proper depression angle with the ground so as to take an image of the ground environment of the farmland suitable for model recognition.
Fig. 2 is a schematic diagram of classification of farmland environment provided by an embodiment of the present invention, and as shown in fig. 2, the embodiment of the present invention provides a method for determining a classification of farmland operation environment with an agricultural machinery dry farmland operation as an application background:
the dry farmland operation environment scenes are firstly classified, and the dry farmland operation environment scenes can be specifically divided into two categories of farmlands and first-class farmlands. The farmland environment is divided into two cases of no crop coverage and crop coverage. The farmland environment can be specifically divided into two conditions of adjacent farmland and adjacent non-farmland according to the adjacent environment as a transition zone of the farmland. For the adjacent farmland type, the three conditions of immature green crop coverage, mature yellow crop coverage and bare soil without crop coverage can be divided according to the difference of ground coverage. Because the adjacent non-farmland type land is relatively complex, the landscape elements can be divided into vegetation coverage type (weeds, shrubs and trees), ridges (bare soil) and artificial facilities according to the structure of the landscape elements.
Through the above analysis of the classification of farmland headers, although the ground cover of some of the header environments is different, such as green crops in adjacent farmland-type headers and green vegetation in adjacent non-farmland-type headers, classification studies can be conducted from the image recognition viewpoint. According to the embodiment of the invention, 6 farmland operation environment categories are divided according to the characteristics of farmland image, such as color, texture and the like, and the image automatic identification research is carried out, wherein the categories are respectively as follows: firstly, no crop farmland image exists; secondly, a crop farmland image is provided; the green vegetation land image comprises two land heads of adjacent green crop coverage and green vegetation coverage of shrubs, weeds and the like; fourthly, the image of the yellow vegetation ground, which comprises the ground covered by mature crops and the ground covered by yellow vegetation such as withered and yellow weeds, shrubs and the like; the image of the head of the bare soil comprises two types of the head of the adjacent farmland, namely the white land and the ridge; sixthly, images of the ground of artificial facilities, such as cement roads, stone slab roads, ditches, fences and the like.
After the farmland operation environment types are determined, the embodiment of the invention provides a farmland identification network model constructed based on deep learning, which is used for learning and analyzing image characteristics such as color, texture and the like of any farmland image inputted in real time so as to output the farmland operation environment types corresponding to the farmland image. In the embodiment of the present invention, the network model framework of the above-described header recognition network model is not specifically limited.
Optionally, after the ground recognition network model is constructed, the image sample of the ground of the farmland with the farmland operation environment category label can be used for pre-training the image sample of the ground of the farmland, and the hyper-parameters of the ground recognition network model are adjusted according to the result of each training until the training result is converged.
The farmland headland automatic identification method provided by the embodiment of the invention is used for identifying the farmland headland image under the agricultural machinery operation scene based on deep learning, can accurately acquire the operation environment of the current agricultural machinery operation, is good in real-time performance, and can provide a feasible technical solution for autonomous perception of the farmland environment of the intelligent agricultural machinery.
Based on the content of the foregoing embodiment, as an optional embodiment, after acquiring the farmland head image, the method further includes: and cutting the farmland head image into an aspect ratio of 1 by taking the short side of the farmland head image as the side length: 1; zooming the first image based on a nearest neighbor interpolation method to obtain a second image; the pixels of the second image are identifiable pixels of the geography-identifying network model.
Generally, the size of the farmland ground image collected in real time is generally larger than the input size of the ground recognition network model, and the farmland ground image is directly input into the ground recognition network model, so that the error of a recognition result is too large, and even the recognition cannot be finished. Therefore, in the farmland headland automatic identification method provided by the embodiment of the invention, after the farmland headland image is acquired, the size of the image needs to be adjusted firstly on the basis of not losing the image characteristics.
Since the length and width of the original image of the farmland image collected in real time are generally different, if the image is directly obtained according to the aspect ratio 1: scaling by 1 can cause the image content of the farmland image to deform, thereby affecting the effect of classification and identification. Therefore, in the embodiment of the invention, the short side of the original image of the farmland head image is taken as the side length, and the image is cut into the length-width ratio of 1: 1 (hereinafter collectively referred to as a first image).
Furthermore, a nearest neighbor interpolation method is adopted, and each first image is zoomed under the condition that image information is ensured not to be lost as much as possible. For example, it may be scaled to 224 × 224 pixels (i.e., recognizable pixels of the ground recognition network model) to meet the pixel input requirements of the ground recognition network model.
The Nearest Neighbor Interpolation (NNI) refers to setting the pixel value of each point in the second image as the Nearest point in the first image. That is, the first image is regarded as a pixel point map a formed by countless pixel points, then a pixel point map b with the same size but more sparse pixel points is covered on the first image, the gray value of each pixel point on the map b is used for searching the gray value of the closest pixel from the map a and assigning the gray value to the point on the map b, and after all points on the map b are assigned, the first image can be reduced to the required multiple, and the pixel point map b is the pixel point map corresponding to the second image.
According to the farmland ground automatic identification method provided by the embodiment of the invention, the farmland ground image is preprocessed before being analyzed by utilizing the pre-constructed ground identification network model, so that the precision of model identification can be effectively improved.
Based on the content of the above embodiment, as an optional embodiment, the heading identification network model is based on MobileNet V2And (4) constructing.
Because the automatic ground identification model needs to be operated on a vehicle-mounted computer of the operating agricultural machine, the operation and storage performance of the automatic ground identification model is lower than that of a desktop computer with the same price. Because the current popular deep learning model consumes huge computer resources and is not suitable for being directly deployed on vehicle-mounted equipment, the ground identification network model needs to be compressed.
The farmland heading automatic identification method provided by the embodiment of the invention uses the MobileNet V with compact model structure design2Said MobileNet V2The method is a light-weight deep neural network provided for embedded equipment such as a mobile phone and the like, and can realize the purpose of reducing the model parameters on the basis of not losing too large prediction precision by using a large number of deep separable convolutions to replace the traditional convolution, so that the operation requirement of a vehicle-mounted computer can be met.
Based on the content of the foregoing embodiment, as an optional embodiment, the network structure of the header identification network model includes an input layer, a first convolution module, an imported Residual Block module, a second convolution module, a global averaging pooling layer, a Dropout layer, a 1 × 1convolution layer, and a Softmax layer, which are connected in sequence; the first convolution module and the second convolution module have the same structure and respectively comprise a convolution layer, a batch normalization layer and a Relu activation layer which are sequentially connected; the invoked Residual Block module comprises 7 invoked Residual blocks which are connected in sequence; shortcuts are used between convolutional layers with a sliding step of 1 in each of the invoked Residual blocks.
FIG. 3 shows a MobileNet V according to an embodiment of the present invention2As shown in fig. 3, an input farmland image (with a pixel of 224 × 224) first passes through a Convolution module (Conv-Block, hereinafter referred to as a first Convolution module) including a Convolution layer (Convolution), a batch normalization layer (BatchNorm), and a Relu activation layer. And then, 7 invoked Residual Block modules are used for reducing the dimension and reducing the parameter number, wherein the convolutional layer with the sliding step length of 2 does not use short cut, and the convolutional layer with the sliding step length of 1 uses short cut. After being output from the inversed Residual Block, the parameters are further reduced through a convolution module and then a Global Average Pooling layer (Global Average potential) is used for replacing a full connection layer. The process is continued with Dropout, 1 × 1Convolution layer (1 × 1 contribution) to prevent overfitting, and finally classified and output through Softmax layer.
Wherein, MobileNet V2The core of (1) is a Separable Convolution (Separable Convolution), which can effectively reduce the parameter number on the premise of sacrificing smaller performance.
The operation steps of each invoked Residual Bloc are as follows: and amplifying the input low-dimensional image features to a high dimension, performing convolution operation in a high-dimension convolution mode, and mapping the low-dimension image features to a low-dimension space by using a linear convolution.
The embodiment of the invention introduces the shortcut idea into the ground-head recognition network model, and aims to solve the problems of gradient divergence and difficult training in the ground-head recognition network model by adding the (authorized) shortcut between the two convolutional layers with the sliding step length of 1.
According to the farmland headland automatic identification method provided by the embodiment of the invention, the lightweight model MobileNet V2 is selected as the prediction model, so that the method is high in accuracy, high in identification speed and high in memory occupation, is supported to be used on a vehicle-mounted computer of an operating agricultural machine, and meets the application requirement of the headland environment perception. The adopted mode based on transfer learning is used for recognizing the farmland headland images, so that a model with good generalization ability can be trained under the condition that the number of farmland headland image samples is limited, and the farmland headland recognition rate under a complex background is greatly improved.
Based on the content of the foregoing embodiment, as an optional embodiment, before inputting the farmland head image to the head recognition network model, the method may further include:
obtaining a plurality of farmland ground image samples of each farmland operation environment type;
cutting the farmland headland image samples according to a preset size proportion by taking the image top of each farmland headland image sample as a cutting starting point in a centrosymmetric mode to obtain multi-size image samples;
turning each farmland ground image sample according to a preset angle to obtain multi-angle image samples;
randomly transforming the brightness, contrast and chromaticity of each farmland head image sample to obtain a multicolor image sample;
constructing a farmland head image sample set consisting of the farmland head image sample, the multi-size image sample, the multi-angle image sample and the multi-color image sample; each image sample in the farmland head image sample set corresponds to a farmland operation environment type label;
and pre-training the ground recognition network model by utilizing the farmland ground image sample set.
Further, after obtaining a plurality of farmland ground image samples of each farmland operation environment type, each image sample in the farmland ground image sample set is also preprocessed, including: and performing cutting processing and scaling processing so that the processed pixel of each image sample is an identifiable pixel of the ground identification network model.
In the embodiment of the present invention, before analyzing the input farmland headland image by using the headland recognition network model, the headland recognition network model needs to be pre-trained, and the whole pre-training process will be described in the following specific embodiment.
Firstly, farmland ground image samples aiming at each farmland operation environment category are collected, 6000 images are counted, and a training sample set is constructed. Wherein, each farmland operation environment type is 1000 respectively. Alternatively, a farmland image sample for each farmland operation environment category can be collected, and a verification sample set and a test sample set are constructed at the same time. The ratio of the number of images in the training sample set, the verification sample set, and the test sample set may be set to 4: 1: 1. the verification sample set is used for verifying the accuracy of the trained ground recognition network model, 1500 images are counted, and the category of each farmland operation environment is 250 images; the test sample set is used for testing the practical application performance of the trained ground identification network model, the total number of the images is 1500, and the category of each farmland operation environment is 250.
Furthermore, image size cutting-scaling and image enhancement preprocessing are respectively carried out on each farmland headland image sample so as to ensure that the obtained image sample meets the input requirement of a headland recognition network model, ensure that the image characteristic information of the image sample is not lost as much as possible, and effectively expand and enhance training set data.
On one hand, because the size of the collected farmland ground image samples is generally larger than the input size of the ground recognition network model, each farmland ground image sample needs to be scaled so as to unify the image sizes.
Since the length and width of the original image of the farmland ground image sample are different, if directly according to the length-width ratio 1: scaling by 1 can cause deformation of image content and influence classification and identification effects. Therefore, in the embodiment of the present invention, the short side of the original image is taken as the side length, and the clipping is performed according to the preset size ratio (for example, the clipping is performed to the image with the aspect ratio of 1: 1).
And on the other hand, the clipped image is zoomed into recognizable pixels of the ground recognition network model by adopting a nearest neighbor interpolation method. And correspondingly scaling the cropped image to 224 x 224 pixels to meet the input requirement of the network and ensure that the image information is not lost as much as possible under the assumption that the recognizable pixels of the ground recognition network model are 224 x 224 pixels.
By adopting the image size cutting-zooming method, one farmland headland image sample can be cut into two pieces, and the problem of data imbalance caused by insufficient quantity of specific categories can be solved.
Furthermore, a multi-scale cutting-scaling method can be adopted to expand the training sample set, for example, the original image is cut and scaled to different scales, so that the farmland boundary characteristics can be more obviously and comprehensively displayed. When the agricultural machinery runs to be close to the ground area, the useful information of the ground appears from the top end, therefore, when the image is cut, the top end of the image of each farmland ground image sample is used as a cutting starting point, the three scales of 60%, 70% and 80% of the original image are respectively cut in a left-right central symmetry mode, and then the three scales are zoomed to the size same as the original image, and 3 multi-size image samples corresponding to the original farmland ground image sample can be obtained.
Furthermore, can also be with every farmland ground image sample is according to presetting the angle and overturning, if to every farmland ground image sample has carried out horizontal turnover once, then can increase one with the multi-angle image sample that farmland ground image sample corresponds.
Further, each multi-size image sample may be subjected to a flipping process to obtain a multi-angle image sample corresponding to each multi-size image sample.
Further, the brightness, contrast and chromaticity of each multi-size image sample can be randomly transformed to eliminate the interference of illumination on the image and obtain a plurality of multi-color image samples.
And finally, forming a farmland over-the-ground image sample set by all the multi-size image samples, the multi-angle image samples, the multi-color image samples and the original images of the farmland over-the-ground image samples.
Further, in the embodiment of the present invention, label labeling may be performed on each image sample in the farmland head image sample set based on a tensrflow deep learning framework. The training sample set, the verification sample set and the test sample set respectively comprise image samples corresponding to the listed 6 farmland operation environment types. Under each image sample folder, image sample data is stored, a TensorFlow framework is used for converting the image sample data into a file with a TFrecord data format, a complete farmland top image sample set to be labeled is constructed, and a data base is laid for model training.
Further, in the farmland headland automatic identification method provided by the embodiment of the invention, in the process of pre-training the headland identification network model, the attenuation learning rate is used for network training, and the value of the learning rate is attenuated, so that the parameter updating speed can be increased in the early stage of training, the network can be ensured not to have too large fluctuation in the later stage of training, and the optimal solution can be more easily approached.
Furthermore, in the process of pre-training the ground recognition network model, a Momentum learning optimization algorithm is used, Momentum of accumulated historical Gradient information is introduced into the algorithm to accelerate (SGD), the problem that the updating amplitude of the SGD optimization algorithm is large in swing can be solved, and convergence to an optimal solution can be accelerated.
Further, in the pre-training process of the ground recognition network model, in order to reduce overfitting during training, L2 regularization is added.
Further, in the process of pre-training the head of the earth recognition network model, exponential moving average is also used for enhancing the generalization capability of the model.
According to the farmland headland automatic identification method provided by the embodiment of the invention, by carrying out multi-scale expansion, image turning and random transformation of image brightness, contrast and chromaticity on farmland headland image samples, not only is a farmland headland image sample set enhanced, but also interference of illumination on the images is favorably eliminated, and the reliability and accuracy of model identification are improved.
Based on the content of the foregoing embodiment, as an optional embodiment, before the pre-training of the headland recognition network model by using the farmland headland image sample set, the method further includes: and pre-training the headland recognition network model by taking ImageNet as a source data set.
Wherein ImageNet is a visual object for use inA large visual database of software studies is identified. Transfer learning is a method for transferring knowledge in one domain (i.e., a source domain) to another domain (i.e., a target domain), so that the target domain can obtain a better learning effect. The embodiment of the invention adopts a transfer learning method, takes ImageNet as a source data set, and carries out source model MobileNet V on the constructed ground recognition network model2And pre-training to obtain a pre-training model, wherein the pre-training model has certain image recognition capability. And then, pre-training the farmland ground image sample set again, and finely adjusting model parameters to enable the model to complete the recognition task of the farmland ground image more quickly.
The farmland headland automatic identification method provided by the embodiment of the invention identifies the farmland headland image based on the transfer learning mode, and identifies the source model MobileNet V in advance2The method comprises the steps of pre-training, and re-pre-training the farmland headland image sample set to the headland recognition network model, so that the model can train a model with good generalization capability under the condition that the quantity of the headland image samples is limited, and the farmland headland recognition rate under the complex background is greatly improved.
The embodiment of the invention provides an automatic farmland heading identification device, which is shown in fig. 4 and comprises a sensor module 1 and a microprocessor module 2. The sensor module 1 is mainly used for acquiring farmland head images; the microprocessor module is loaded with a ground identification network model and is used for receiving the farmland ground image so as to output the farmland operation environment category corresponding to the farmland ground image; the farmland heading identification network model is obtained by training farmland heading image samples with farmland operation environment class labels; the farmland operation environment categories comprise non-crop farmlands, green vegetation lands, yellow vegetation lands, bare soil lands and artificial facilities lands.
Fig. 5 is a schematic flow chart of another automatic farmland headland recognition method provided by the embodiment of the present invention, and as shown in fig. 5, the whole flow of the automatic farmland headland recognition device provided by the embodiment of the present invention during working may include three parts, namely data preparation, model training, and model deployment.
In the data preparation stage, the method mainly comprises the steps of collecting farmland ground images so as to obtain a farmland ground image sample set containing a certain number of various farmland operation environment types, and performing image data enhancement processing on the collected farmland ground image sample set by means of cutting-zooming, image enhancement and the like so as to provide a more sufficient training sample set, a verification sample set and a testing sample set.
In the model training stage, MobileNet V with compact model structural design is selected firstly2A headland recognition network model is constructed, and then the method of transfer learning is adopted, ImageNet is used as a source data set to finish the MobileNet V2And after pre-training, pre-training the ground recognition network model by utilizing the constructed farmland ground image sample set.
Furthermore, after the model is pre-trained by using each farmland headland image sample, the pre-trained headland recognition network model can be verified by using a verification sample set, and the hyper-parameters of the model are properly adjusted according to the verification result so as to realize the optimization of the model until the pre-training result is converged, and the pre-training process of the whole model is completed, so that the optimal headland recognition network model is obtained.
In the model deployment stage, the microprocessor module loaded with the ground identification network model can be deployed in an agricultural vehicle-mounted device, so that the agricultural vehicle-mounted device is utilized to realize the identification of the farmland ground environment.
It should be noted that, in the embodiment of the present invention, in order to accurately verify MobileNet V2The actual classification accuracy of the network is realized by using a test sample set to perform pre-training on the MobileNet V2A test of network generalization capability is carried out, and a confusion matrix and F1-score are used for evaluating the accuracy of model identification.
In the image recognition accuracy evaluation, the accuracy, recall and F1-score can be further calculated by the confusion matrix. FIG. 6 is a visual confusion matrix diagram of the farmland automatic identification result provided by the embodiment of the invention; as shown in FIG. 6, each column in the matrix represents a prediction category, and the total number of each column represents the number of data predicted for that category; each row represents a true attribution category of data, and the total number of data in each row represents the number of data instances for that category.
Precision (Precision) refers to the proportion of the number of cases correctly predicted as positive by the model to the number of cases correctly predicted as positive by all models. Recall (Recall) is the proportion of the number of cases correctly predicted by the model to be positive cases to the number of cases that are all actually positive cases. F1-score can be regarded as a harmonic mean of model accuracy and recall, with a maximum of 1 and a minimum of 0, and is calculated as:
table 1 shows the accuracy, recall rate and F1-score results of the automatic farmland heading recognition apparatus provided by the embodiment of the present invention for recognizing heading images of 6 farmland work environment types.
TABLE 1
As shown in Table 1, MobileNet V2The results of the analyses performed on the test sample set showed a Precision average of 0.97 and a Recall average of 0.97, based on MobileNet V2The constructed ground identification network model can accurately identify each type of sample; the average value of the F1-score is 0.97, which shows that the land recognition network model can accurately recognize 6 farmland land types under the natural environment, has good robustness and robustness, and can meet the requirements of practical application.
Fig. 7 is another automatic farmland headland recognition device provided by an embodiment of the present invention, and as shown in fig. 7, the device integrates functional modules such as an input module, an output module, a storage module, and a sensor module, with a microprocessor module as a core, and constructs an embedded terminal device suitable for an agricultural machine vehicle-mounted scene application, so as to implement automatic farmland headland environment recognition.
The micro-processing module is the core of the whole device control and is also a key module for running the ground identification network model. The input module can be selected from a keyboard, a touch screen and other equipment and is mainly used for interaction of a human-computer interface so as to realize configuration and operation of farmland heading identification application. The output module can select an LCD display screen for displaying a real-time image of the running front of the operating agricultural machine and outputting a ground type identification result. The storage module can be a hard disk or a flash memory card and is used for storing the model and the acquired head-of-the-earth environment image. The sensor module is used for image acquisition, can select the camera of high resolution, shoots the image data in operation agricultural machinery before going in real time to input for microprocessor module through the interface.
Specifically, the whole process of recognizing the farmland headland image by the automatic farmland headland recognition device can be as follows:
the training model under Tensorflow is saved as h5 format and then converted to tflite format model. Loading a ground identification network model through a Tensorflow Lite tool, configuring image prediction information while loading the model, acquiring an input layer and an input layer of the network, preprocessing and predicting farmland images acquired by a camera in real time, displaying a maximum probability prediction label, and realizing automatic identification of the farmland ground images.
According to the automatic farmland headland recognition device provided by the embodiment of the invention, based on the modularized integration concept of an embedded hardware technology and based on an embedded mainboard and functional modules such as an integrated microprocessor, an input sensor, an output sensor, a storage sensor and an image acquisition sensor, a vehicle-mounted embedded terminal device is constructed, is installed on agricultural machinery, acquires a farmland image in front of driving in real time, and realizes automatic recognition of a headland environment through a deployed farmland headland image recognition model.
It should be noted that, when being specifically executed, the automatic farmland headland recognition device provided in the embodiment of the present invention may be implemented based on the automatic farmland headland recognition method described in any of the above embodiments, and details of this embodiment are not described herein.
Fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, and as shown in fig. 8, the electronic device may include: a processor (processor)810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call the logic instructions in the memory 830 to execute the method for automatically recognizing the farmland heading, which mainly comprises: obtaining a farmland head image; inputting the farmland head image into the head recognition network model, and obtaining a farmland operation environment category corresponding to the farmland head image according to an output result of the head recognition network model; the farmland heading network model is obtained by training farmland heading image samples with farmland operation environment class labels; the farmland operation environment categories comprise non-crop farmland, green vegetation land, yellow vegetation land, bare soil land and artificial facility land.
In addition, the logic instructions in the memory 830 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and the like.
In addition, the logic instructions in the memory 830 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, an embodiment of the present invention further provides a computer program product, where the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, where when the program instructions are executed by a computer, the computer is capable of executing the automatic farmland heading identification method provided by the above-mentioned method embodiments, and mainly includes: obtaining a farmland head image; inputting the farmland head image into the head recognition network model, and obtaining a farmland operation environment category corresponding to the farmland head image according to an output result of the head recognition network model; the farmland heading network model is obtained by training farmland heading image samples with farmland operation environment class labels; the farmland operation environment categories comprise non-crop farmland, green vegetation land, yellow vegetation land, bare soil land and artificial facility land.
In another aspect, an embodiment of the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program is implemented to perform the method for automatically identifying a farmland heading provided in the foregoing embodiments when executed by a processor, and the method mainly includes: obtaining a farmland head image; inputting the farmland head image into the head recognition network model, and obtaining a farmland operation environment category corresponding to the farmland head image according to an output result of the head recognition network model; the farmland heading network model is obtained by training farmland heading image samples with farmland operation environment class labels; the farmland operation environment categories comprise non-crop farmland, green vegetation land, yellow vegetation land, bare soil land and artificial facility land.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A farmland heading automatic identification method is characterized by comprising the following steps:
obtaining a farmland head image;
inputting the farmland head image into a head identification network model, and obtaining a farmland operation environment category corresponding to the farmland head image according to an output result of the head identification network model;
the farmland heading identification network model is obtained by training farmland heading image samples with farmland operation environment class labels;
the farmland operation environment categories comprise non-crop farmlands, green vegetation lands, yellow vegetation lands, bare soil lands and artificial facilities lands.
2. The method for automatically identifying the farmland heading as claimed in claim 1, further comprising, after acquiring the image of the farmland heading:
and cutting the farmland head image into an aspect ratio of 1 by taking the short side of the farmland head image as the side length: 1;
zooming the first image based on a nearest neighbor interpolation method to obtain a second image; the pixels of the second image are identifiable pixels of the geography-identifying network model.
3. The method of claim 1, wherein the headland recognition network model is based on MobileNet V2And (4) constructing.
4. The farmland automatic ground identification method according to claim 3, wherein the network structure of the ground identification network model comprises an input layer, a first convolution module, an Inverted Residual Block module, a second convolution module, a global averaging pooling layer, a Dropout layer, a 1 x 1convolution layer and a Softmax layer which are connected in sequence;
the first convolution module and the second convolution module have the same structure and respectively comprise a convolution layer, a batch normalization layer and a Relu activation layer which are sequentially connected;
the invoked Residual Block module comprises 7 invoked Residual blocks which are connected in sequence;
shortcuts are used between convolutional layers with a sliding step of 1 in each of the invoked Residual blocks.
5. The method of claim 1, wherein before inputting the farmland headland image into the headland recognition network model, further comprising:
obtaining a plurality of farmland ground image samples of each farmland operation environment type;
cutting the farmland headland image samples according to a preset size proportion by taking the image top of each farmland headland image sample as a cutting starting point in a centrosymmetric mode to obtain multi-size image samples;
turning each farmland ground image sample according to a preset angle to obtain multi-angle image samples;
randomly transforming the brightness, contrast and chromaticity of each farmland head image sample to obtain a multicolor image sample;
constructing a farmland head image sample set consisting of the farmland head image sample, the multi-size image sample, the multi-angle image sample and the multi-color image sample; each image sample in the farmland head image sample set corresponds to a farmland operation environment type label;
and pre-training the ground recognition network model by utilizing the farmland ground image sample set.
6. The method of claim 5, wherein after obtaining a plurality of farmland image samples for each farmland operation environment category, preprocessing each image sample in the farmland image sample set further comprises: and performing cutting processing and scaling processing so that the processed pixel of each image sample is an identifiable pixel of the ground identification network model.
7. The method of claim 5, further comprising, before pre-training the model of the farmland headland recognition network with the set of farmland headland image samples, the steps of:
and pre-training the headland recognition network model by taking ImageNet as a source data set.
8. The utility model provides a farmland ground automatic identification equipment which characterized in that includes: a sensor module and a microprocessor module;
the sensor module is used for acquiring a farmland head image;
the microprocessor module is loaded with a ground identification network model and is used for receiving the farmland ground image so as to output the farmland operation environment category corresponding to the farmland ground image;
the farmland heading identification network model is obtained by training farmland heading image samples with farmland operation environment class labels;
the farmland operation environment categories comprise non-crop farmlands, green vegetation lands, yellow vegetation lands, bare soil lands and artificial facilities lands.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the method for automatic farmland headland recognition according to any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for automatic farmland headland farmland heading identification according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011287841.7A CN112464766B (en) | 2020-11-17 | Automatic farmland land identification method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011287841.7A CN112464766B (en) | 2020-11-17 | Automatic farmland land identification method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112464766A true CN112464766A (en) | 2021-03-09 |
| CN112464766B CN112464766B (en) | 2024-07-02 |
Family
ID=
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113052369A (en) * | 2021-03-15 | 2021-06-29 | 北京农业智能装备技术研究中心 | Intelligent agricultural machinery operation management method and system |
| CN113469116A (en) * | 2021-07-20 | 2021-10-01 | 河南科技大学 | Face expression recognition method combining LBP (local binary pattern) features and lightweight neural network |
| CN113778110A (en) * | 2021-11-11 | 2021-12-10 | 山东中天宇信信息技术有限公司 | Intelligent agricultural machine control method and system based on machine learning |
| CN114067158A (en) * | 2021-11-17 | 2022-02-18 | 江苏天汇空间信息研究院有限公司 | Farmland use state monitoring system and method applying multi-source remote sensing data |
| CN114485612A (en) * | 2021-12-29 | 2022-05-13 | 广州极飞科技股份有限公司 | Route generation method and device, unmanned working vehicle, electronic device and storage medium |
| CN114972807A (en) * | 2022-05-17 | 2022-08-30 | 北京百度网讯科技有限公司 | Method and device for determining image recognition accuracy, electronic equipment and medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102167038A (en) * | 2010-12-03 | 2011-08-31 | 北京农业信息技术研究中心 | Method and device for generating all-region-covering optimal working path for farmland plot |
| CN108776803A (en) * | 2018-04-20 | 2018-11-09 | 中国农业大学 | The method and system of weeds in a kind of removal farmland |
| US20180373932A1 (en) * | 2016-12-30 | 2018-12-27 | International Business Machines Corporation | Method and system for crop recognition and boundary delineation |
| CN110825118A (en) * | 2019-12-23 | 2020-02-21 | 河南大学 | Multi-unmanned aerial vehicle cooperative farmland spraying method based on deep learning algorithm |
| CN111259898A (en) * | 2020-01-08 | 2020-06-09 | 西安电子科技大学 | Crop segmentation method based on unmanned aerial vehicle aerial image |
| US20200193589A1 (en) * | 2018-12-10 | 2020-06-18 | The Climate Corporation | Mapping field anomalies using digital images and machine learning models |
| CN111724340A (en) * | 2020-05-09 | 2020-09-29 | 北京农业智能装备技术研究中心 | Farmland margin line visual detection method and system |
| CN111814597A (en) * | 2020-06-20 | 2020-10-23 | 南通大学 | Urban function partitioning method coupling multi-label classification network and YOLO |
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102167038A (en) * | 2010-12-03 | 2011-08-31 | 北京农业信息技术研究中心 | Method and device for generating all-region-covering optimal working path for farmland plot |
| US20180373932A1 (en) * | 2016-12-30 | 2018-12-27 | International Business Machines Corporation | Method and system for crop recognition and boundary delineation |
| CN108776803A (en) * | 2018-04-20 | 2018-11-09 | 中国农业大学 | The method and system of weeds in a kind of removal farmland |
| US20200193589A1 (en) * | 2018-12-10 | 2020-06-18 | The Climate Corporation | Mapping field anomalies using digital images and machine learning models |
| CN110825118A (en) * | 2019-12-23 | 2020-02-21 | 河南大学 | Multi-unmanned aerial vehicle cooperative farmland spraying method based on deep learning algorithm |
| CN111259898A (en) * | 2020-01-08 | 2020-06-09 | 西安电子科技大学 | Crop segmentation method based on unmanned aerial vehicle aerial image |
| CN111724340A (en) * | 2020-05-09 | 2020-09-29 | 北京农业智能装备技术研究中心 | Farmland margin line visual detection method and system |
| CN111814597A (en) * | 2020-06-20 | 2020-10-23 | 南通大学 | Urban function partitioning method coupling multi-label classification network and YOLO |
Non-Patent Citations (2)
| Title |
|---|
| 李云伍;徐俊杰;刘得雄;于尧;: "基于改进空洞卷积神经网络的丘陵山区田间道路场景识别", 农业工程学报, no. 07, 8 April 2019 (2019-04-08) * |
| 薛金林;闫嘉;范博文;: "多类农田障碍物卷积神经网络分类识别方法", 农业机械学报, no. 1, 16 November 2018 (2018-11-16) * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113052369A (en) * | 2021-03-15 | 2021-06-29 | 北京农业智能装备技术研究中心 | Intelligent agricultural machinery operation management method and system |
| CN113052369B (en) * | 2021-03-15 | 2024-05-10 | 北京农业智能装备技术研究中心 | Intelligent agricultural machinery operation management method and system |
| CN113469116A (en) * | 2021-07-20 | 2021-10-01 | 河南科技大学 | Face expression recognition method combining LBP (local binary pattern) features and lightweight neural network |
| CN113778110A (en) * | 2021-11-11 | 2021-12-10 | 山东中天宇信信息技术有限公司 | Intelligent agricultural machine control method and system based on machine learning |
| CN114067158A (en) * | 2021-11-17 | 2022-02-18 | 江苏天汇空间信息研究院有限公司 | Farmland use state monitoring system and method applying multi-source remote sensing data |
| CN114485612A (en) * | 2021-12-29 | 2022-05-13 | 广州极飞科技股份有限公司 | Route generation method and device, unmanned working vehicle, electronic device and storage medium |
| CN114485612B (en) * | 2021-12-29 | 2024-04-26 | 广州极飞科技股份有限公司 | Route generation method and device, unmanned operation vehicle, electronic equipment and storage medium |
| CN114972807A (en) * | 2022-05-17 | 2022-08-30 | 北京百度网讯科技有限公司 | Method and device for determining image recognition accuracy, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111986099B (en) | Tillage monitoring method and system based on convolutional neural network with residual error correction fused | |
| Jia et al. | Detection and segmentation of overlapped fruits based on optimized mask R-CNN application in apple harvesting robot | |
| CN110059694B (en) | Intelligent identification method for character data in complex scene of power industry | |
| CN109960742B (en) | Local information searching method and device | |
| CN113065558A (en) | Lightweight small target detection method combined with attention mechanism | |
| CN110163213B (en) | Remote sensing image segmentation method based on disparity map and multi-scale depth network model | |
| CN109446970A (en) | A kind of Intelligent Mobile Robot road scene recognition methods based on deep learning | |
| CN111126258A (en) | Image recognition method and related device | |
| CN108090447A (en) | Hyperspectral image classification method and device under double branch's deep structures | |
| CN113159300A (en) | Image detection neural network model, training method thereof and image detection method | |
| CN111062441A (en) | Scene classification method and device based on self-supervision mechanism and regional suggestion network | |
| CN114187450A (en) | Remote sensing image semantic segmentation method based on deep learning | |
| CN109492665A (en) | Detection method, device and the electronic equipment of growth period duration of rice | |
| CN112419333B (en) | Remote sensing image self-adaptive feature selection segmentation method and system | |
| CN104063686A (en) | System and method for performing interactive diagnosis on crop leaf segment disease images | |
| CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
| CN114972646B (en) | Method and system for extracting and modifying independent ground objects of live-action three-dimensional model | |
| CN113269182A (en) | Target fruit detection method and system based on small-area sensitivity of variant transform | |
| CN114943893A (en) | Feature enhancement network for land coverage classification | |
| CN114972208A (en) | YOLOv 4-based lightweight wheat scab detection method | |
| CN115115954A (en) | Intelligent identification method for pine nematode disease area color-changing standing trees based on unmanned aerial vehicle remote sensing | |
| CN111914596B (en) | Lane line detection method, device, system and storage medium | |
| CN114841961A (en) | Wheat scab detection method based on image enhancement and improvement of YOLOv5 | |
| CN113033386B (en) | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system | |
| CN113673369A (en) | Remote sensing image scene planning method and device, electronic equipment and storage medium |
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 | ||
| GR01 | Patent grant |