CN116868974A - Accurate medicine weeding device that spouts based on weeds kind - Google Patents
Accurate medicine weeding device that spouts based on weeds kind Download PDFInfo
- Publication number
- CN116868974A CN116868974A CN202310818917.1A CN202310818917A CN116868974A CN 116868974 A CN116868974 A CN 116868974A CN 202310818917 A CN202310818917 A CN 202310818917A CN 116868974 A CN116868974 A CN 116868974A
- Authority
- CN
- China
- Prior art keywords
- weed
- image
- precision
- weeds
- network
- 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
- 241000196324 Embryophyta Species 0.000 title claims abstract description 142
- 239000003814 drug Substances 0.000 title claims abstract description 58
- 238000009333 weeding Methods 0.000 title claims abstract description 35
- 238000005507 spraying Methods 0.000 claims abstract description 78
- 239000000575 pesticide Substances 0.000 claims abstract description 42
- 239000004009 herbicide Substances 0.000 claims abstract description 18
- 230000002363 herbicidal effect Effects 0.000 claims abstract description 10
- 239000007921 spray Substances 0.000 claims description 31
- 238000000034 method Methods 0.000 claims description 29
- 238000012549 training Methods 0.000 claims description 24
- 230000008569 process Effects 0.000 claims description 21
- 238000001514 detection method Methods 0.000 claims description 17
- 230000000694 effects Effects 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 14
- 239000007788 liquid Substances 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 12
- 238000002372 labelling Methods 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 6
- 239000003086 colorant Substances 0.000 claims description 3
- 238000005286 illumination Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 229940079593 drug Drugs 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 4
- 206010059866 Drug resistance Diseases 0.000 description 3
- 241001290610 Abildgaardia Species 0.000 description 2
- 241000209504 Poaceae Species 0.000 description 2
- 238000012271 agricultural production Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000003912 environmental pollution Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000002689 soil Substances 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 241000238631 Hexapoda Species 0.000 description 1
- 241000607479 Yersinia pestis Species 0.000 description 1
- 230000001580 bacterial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000035558 fertility Effects 0.000 description 1
- 239000003337 fertilizer Substances 0.000 description 1
- 239000010916 herbicide waste Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 235000015097 nutrients Nutrition 0.000 description 1
- 230000029553 photosynthesis Effects 0.000 description 1
- 238000010672 photosynthesis Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01M—CATCHING, TRAPPING OR SCARING OF ANIMALS; APPARATUS FOR THE DESTRUCTION OF NOXIOUS ANIMALS OR NOXIOUS PLANTS
- A01M7/00—Special adaptations or arrangements of liquid-spraying apparatus for purposes covered by this subclass
- A01M7/0025—Mechanical sprayers
- A01M7/0032—Pressure sprayers
- A01M7/0042—Field sprayers, e.g. self-propelled, drawn or tractor-mounted
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01M—CATCHING, TRAPPING OR SCARING OF ANIMALS; APPARATUS FOR THE DESTRUCTION OF NOXIOUS ANIMALS OR NOXIOUS PLANTS
- A01M7/00—Special adaptations or arrangements of liquid-spraying apparatus for purposes covered by this subclass
- A01M7/0089—Regulating or controlling systems
Landscapes
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Insects & Arthropods (AREA)
- Pest Control & Pesticides (AREA)
- Wood Science & Technology (AREA)
- Zoology (AREA)
- Environmental Sciences (AREA)
- Mechanical Engineering (AREA)
- Catching Or Destruction (AREA)
Abstract
The invention discloses an accurate pesticide spraying weeding device based on weed species, which is characterized by comprising: the system comprises an image acquisition system, a weed identification and classification system, a pesticide spraying control and decision system, a target spraying weeding system and a trolley travelling system; the image acquisition system is fixed on a U-shaped frame at the front end of the trolley traveling system, and the weed identification and classification system is fixed at the front end of the chassis of the trolley traveling system. The invention relates to the technical field of plant protection machinery and pesticide application, in particular to an accurate pesticide spraying weeding device based on weed types. The invention aims to solve the technical problem of providing the accurate pesticide spraying weeding device based on the weed types, which can realize the accurate spraying of different types of herbicides according to the captured image types of farmland weeds, can ensure that the pesticide utilization rate is higher, reduce the pollution and damage of the herbicide to the farmland ecological environment, and can also effectively inhibit the harm of drug-resistant weeds.
Description
Technical Field
The invention relates to the technical field of plant protection machinery and pesticide application, in particular to an accurate pesticide spraying weeding device based on weed types.
Background
Weeds in farmlands are a key factor affecting crop growth, and in the growth process of crops, weeds not only fight water, fight light, fight fertilizer, fight growth space, other nutrients and the like with crops, but also serve as hosts for a plurality of diseases and insect pests, thereby affecting photosynthesis of the crops and growth of the crops, and causing serious loss of agricultural production. For weed control, at present, a method for spraying pesticides on average is still commonly used in China, namely, the same type of pesticides are sprayed on a piece of land with the same crop and weed extent, but only a small part of pesticides can act on targets in the pesticide spraying process, and a large amount of pesticides can drop into the soil, so that the pesticide utilization rate is low, the herbicide is wasted, the soil fertility is lost and the environment is polluted; in addition, because of the long-term and large-scale use of herbicides, the drug resistance of weeds is enhanced, and the statistics shows that 513 organisms with 267 weeds generate drug resistance to 21 herbicides in various farmland systems, the rapid evolution of herbicide-resistant weed species enables the research of non-chemical methods to be revived, so that the change of the traditional chemical weeding use mode and the improvement of the application technology become key problems for improving and developing green agriculture and accurate agriculture.
In order to solve the problems of herbicide waste, environmental pollution and the like caused by average pesticide spraying. The variable spraying technology integrating advanced technologies such as machine vision, electromechanical integration, low-quantity spraying, full hydraulic driving and the like is used abroad, and the spraying machine is mainly characterized by wide and long width, and is suitable for large-area farmland abroad, high operation precision and high operation efficiency. The variable pesticide spraying technology reduces the harm of pesticides to the environment and crops, improves the production efficiency of agriculture, and reduces the pesticide cost in the pesticide spraying process. Many variable spraying machines are introduced in China, and the medium-sized and small-sized variable spraying machines are mainly designed for being suitable for the topography of China and carrying out structural improvement. At present, the domestic variable spraying machine mainly comprises a variable spraying system based on machine vision, a variable spraying system based on a prescription diagram, an intelligent spraying system for mixing medicines in real time and the like, but the spraying technologies still spray the same type of herbicide to weeds in the same farmland, so that the weeds generate drug resistance due to the use of the herbicide for tired months, and the weeding effect is obviously reduced. Therefore, developing high-efficiency intelligent weeding equipment classifies weeds in the same farmland and controls a pesticide spraying system to spray herbicides of different types, the premise of accurately spraying the target pesticide and automatically weeding by machinery is to accurately identify crops and weeds, an on-line weed detection and classification robot based on vision is a key factor for specially treating single weeds, and an actuating mechanism and a control system of the weeding machine are important supports.
At present, the concept and theory of precision agriculture have been widely accepted internationally, and a great deal of research on precision agriculture has been underway. The precise spraying technology is one of precise agricultural technologies, two methods for implementing precise spraying are mainly adopted, one method is to spray targets, and more precise spraying is carried out according to the characteristics of a spraying object; the other is variable spraying, and the variable spraying can control the spraying quantity so as to prevent excessive spraying or waste of the medicament. Through the precise spraying technology, the spraying effect can be improved, the pesticide usage amount can be saved, the crop quality can be improved, the environmental pollution can be reduced, the agricultural production efficiency can be improved, and the like. The target spraying and the variable spraying are two spraying technologies commonly used in modern agriculture, and the two spraying methods can improve the utilization efficiency of pesticides, reduce the waste of the pesticides and the pollution to the environment, and can prevent the excessive use of the pesticides to a certain extent.
Disclosure of Invention
The invention aims to solve the technical problem of providing the accurate pesticide spraying weeding device based on the weed types, which can realize the accurate spraying of different types of herbicides according to the captured image types of farmland weeds, can ensure that the pesticide utilization rate is higher, reduce the pollution and damage of the herbicide to the farmland ecological environment, and can also effectively inhibit the harm of drug-resistant weeds.
The invention adopts the following technical scheme to realize the aim of the invention:
accurate medicine weeding device that spouts based on weeds kind, its characterized in that includes: the system comprises an image acquisition system, a weed identification and classification system, a pesticide spraying control and decision system, a target spraying weeding system and a trolley travelling system; the system comprises an image acquisition system, a weed identification and classification system, a pesticide spraying control and decision system and a weed identification and classification system, wherein the image acquisition system is fixed on a U-shaped frame at the front end of the trolley traveling system; the image acquisition system is electrically connected with the weed identification and classification system, and the weed identification and classification system and the target spraying weeding system are respectively electrically connected with the pesticide spraying control and decision system.
As a further limitation of this technical scheme, to target spraying weeding system includes a set of medical kit, a set of medical kit sets up side by side dolly traveling system chassis rear end, medical kit fixed connection dolly traveling system's chassis, the medical kit is provided with the medicine outlet, the medical kit corresponds go out medicine outlet fixed connection return pipe way one end, return pipe way one end intercommunication the medical kit, return pipe way other end fixed intercommunication the medical kit, install filter, medicine pump, solenoid valve and relief valve on the return pipe way, every solenoid valve fixed connection respectively a set of shower nozzle, every the shower nozzle communicates respectively and corresponds the solenoid valve.
As a further limitation of the technical scheme, the image acquisition system sends the acquired weed images to the weed identification and classification system, the weed identification and classification system analyzes and processes the acquired weed images, then sends the identified weed types to the pesticide spraying control and decision system, and finally the pesticide spraying control and decision system sprays the herbicide.
As a further limitation of the present technical solution, the workflow of the weed identification and classification system is:
s1: the method comprises the steps of collecting a data set, wherein the height of a shot image is about 30-50 cm when weeds are collected in the early stage, and the angle is about 60-90 degrees, namely, the shooting angle is close to the vertical weeds, and the recognition accuracy of the follow-up weeds can be influenced due to different weather, illumination and other conditions, so that the weed images are collected on sunny days and cloudy days respectively;
s2: labeling the pictures into an xml file in a VOC format by using labeling software LabelImg, labeling weeds by using a rectangular frame, wherein the frame displays the name and position information of the picture labeling, and before the pictures are input into a convolution network to extract information, carrying out image normalization processing, image normalization processing and image feature extraction on the acquired pictures;
s21: carrying out image normalization; the range of the pixel points of the acquired RGB image is between 0 and 255, and in order to accelerate network training, the pixel is converted into between 0 and 1 by using a formula (1) to perform normalization operation;
wherein: x is x i Pixel value, X, representing the i-th point in the image max And X min Representing the maximum and minimum values of the image pixels, respectively;
s22: performing image standardization processing; because the data information of the image is distributed more dispersedly, the difficulty of network training is increased, and the image standardization is needed, the main principle is that the data is subjected to centering treatment through mean removal, the mean value and variance of the pixel points are obtained through formulas (2) and (3), and then the standardization treatment is carried out through formula (4);
wherein: μ is the mean of the image pixels, σ represents the standard deviation of the pixels, N represents the number of pixels of the image x;
s23: extracting image features; in the process of feature extraction, low-level features such as textures, frames and colors are generally extracted by a shallow network, high-level and abstract features are extracted by a deep network, and the main purpose is to fit a picture target with a marked real area so as to achieve the minimum error rate; extracting the characteristic information of the neural network from the processed picture through a formula (5);
wherein: z (u, v) represents the pixel value of the image in the v column of the u th row corresponding to the characteristic image after convolution, and x i，j Representing the pixel value of the image of the ith row and the jth column corresponding to the convolved image, and k represents the convolution kernel;
s3: inputting the picture into a convolution network to extract information; the size of the image after convolution operation is changed, mainly the number of channels is increased and the length and width are changed; the neural network training process is a process of continuously updating parameter fitting training data back and forth, and in order to improve the network optimization speed, the feature map and the performance are subjected to standardized processing, so that the convergence capacity of the network is enhanced;
after the network training is finished, a better weight value can be obtained among all layers, the optimal prediction effect can be achieved, and the weight is reserved for detecting weeds in the follow-up process; based on the previously obtained network weight, the picture obtained in the actual environment is processed and then is transmitted into the network, because the weight parameters among layers reach a better level in the training process, the input image only needs to be transmitted forward under the parameters, and the actual target class probability is finally obtained;
wherein: l represents the loss function, p, of the model i A set of data, t, representing model predictions i Representing data corresponding to a real sample of a model, N cls The predicted sample size, i, represents the ith bit data value, L in the array cls Model classification loss function, λ represents weight, N reg Number of samples regressed, L reg Model regression loss function,Representing the category of the prediction->Representing the true category;
adopting precision P, recall rate R and average precision mean value m AP As the evaluation index of target detection, the three values are all in the range of 0,1]At the same time introduce F 1 Carrying out harmonic mean evaluation on the values;
the precision represents the ratio of the number of a certain type of weeds detected correctly to the number predicted as the type of weeds, and the recall rate represents the ratio of all targets of the type of weeds in the sample to be predicted correctly;
wherein: t (T) P Representing the number of correctly divided positive samples;
F P indicating the number of erroneous divisions into positive samples;
T N representing the number of correctly divided negative samples;
F N indicating the number of false divisions into negative samples.
As a further limitation of the technical scheme, the average precision represents the detection effect of the used detection network model on a certain class of targets, the larger the value is, the better the overall detection effect is, otherwise, the worse the overall detection effect is, the average precision is mainly represented by a precision and recall curve, namely a PR curve, the abscissa of the curve is that the recall reflects the covering capability of an aligned sample, the precision of a predicted positive sample is reflected by the precision of a vertical axis, and the calculation of the average precision is taken as the integral of the precision and recall curve on [0,1 ];
the average precision mean value represents the mean value of the average precision of all the categories in the data set, and the calculation method is the ratio of the sum of the average precision of all the categories to the number of all the categories;
wherein: a is that p Representing average precision, R representing recall, m AP Represents the average precision mean value, n represents the category number of all weeds;
F 1 the value is a comprehensive evaluation index based on precision and recall, which is defined as a harmonic mean of precision and recall;
the weed identification result is displayed on the computer end in real time, and the identified weed type and the identified probability are fed back according to the performance of the pre-training network structure.
As a further limitation of the technical scheme, the pesticide spraying control and decision system can judge the existence of weeds and the types of the weeds according to different types of weed information fed back by the weed identification and classification system, send corresponding control instructions according to the identification result, and open the corresponding electromagnetic valves to spray the corresponding types of herbicides.
As a further limitation of the present technical solution, the image acquisition system is an industrial vision camera.
As a further limitation of the technical scheme, a liquid level sensor is fixedly connected in the medicine chest, and the liquid level sensor is electrically connected with the medicine spraying control and decision system.
As a further limitation of the technical scheme, the medicine box is provided with a medicine inlet.
Compared with the prior art, the invention has the advantages and positive effects that:
the accurate pesticide spraying weeding device based on the weed types can judge whether the weeds exist or not according to the real-time identification result, the weed types are identified through the weed identification and classification system, the pesticide spraying control and decision system sends corresponding control instructions according to the identification result, the corresponding electromagnetic valves are opened, corresponding medicaments are sprayed on target weeds, and accurate spraying of the weeds is achieved.
Drawings
FIG. 1 is a diagram showing the basic constitution of 4 herbicidal spray systems according to the present invention based on weed classification.
FIG. 2 is a schematic view of a portion of an image required by the image recognition system of the present invention to recognize types of 3 weeds.
FIG. 3 is a schematic diagram of a rectangular frame of a picture using LabelImg labeling software according to the present invention.
Fig. 4 is a schematic diagram of the invention for labeling weed pictures in xml format.
Fig. 5 is a schematic diagram of the result of image recognition at the pre-training stage of the present invention.
FIG. 6 is a schematic diagram of the overall structure of each system of the present invention.
FIG. 7 is a schematic view of the nozzle mounting location of the present invention.
Fig. 8 is a schematic structural flow diagram of the whole present invention.
FIG. 9 is a schematic view of a portion of a piping structure according to the present invention.
In the figure: 1. the device comprises a nozzle, a medicine outlet, a medicine liquid box, a medicine inlet, a liquid level sensor, a trolley running system, a medicine spraying control and decision system, a weed identification and classification system, an image acquisition system, a filter, a medicine liquid pump, a solenoid valve, a safety valve and a safety valve.
Detailed Description
One embodiment of the present invention will be described in detail below with reference to the attached drawings, but it should be understood that the scope of the present invention is not limited by the embodiment.
The invention comprises the following steps: the system comprises an image acquisition system 9, a weed identification and classification system 8, a pesticide spraying control and decision system 7, a target spraying weeding system and a trolley travelling system 6;
the image acquisition system 9 is fixed on a U-shaped frame at the front end of the trolley traveling system 6, the weed identification and classification system 8 is fixed at the front end of a chassis of the trolley traveling system 6, and the pesticide spraying control and decision system 7 is arranged at the rear side of the weed identification and classification system 8 and is fixedly connected with the trolley traveling system 6;
the image acquisition system 9 is electrically connected with the weed identification and classification system 8, and the weed identification and classification system 8 and the target spraying weeding system are respectively electrically connected with the spraying control and decision system 7.
The targeted spraying weeding system comprises a group of medicine boxes 3, the medicine boxes 3 are arranged side by side at the rear end of a chassis of the trolley traveling system 6, the medicine boxes 3 are fixedly connected with the chassis of the trolley traveling system 6, the medicine boxes 3 are provided with medicine outlets 2, the medicine boxes 3 correspond to the medicine outlets 2, one ends of loop pipes are fixedly connected with the medicine boxes 3, the other ends of the loop pipes are fixedly connected with the medicine boxes 3, a filter 10, a medicine pump 11, an electromagnetic valve 12 and a safety valve 13 are arranged on the loop pipes, each electromagnetic valve 12 is fixedly connected with a group of spray heads 1, and each spray head 1 is respectively connected with the corresponding electromagnetic valve 12.
For example, there are three solenoid valves 12, each solenoid valve 12 is fixedly connected with three spray heads 1, each group of first spray heads 1 is communicated with the first solenoid valve 12, each group of second spray heads 1 is communicated with the second solenoid valve 12, and each group of third spray heads 1 is communicated with the third solenoid valve 12.
The liquid medicine flows from the medicine box through the medicine outlet 2, flows through the pipeline filter 10 and then flows into the water inlet of the medicine pump 11, the medicine pump pumps out two pipelines in total, one pipeline flows through the safety valve 13 and flows back to the medicine box through the safety valve branch, and the other branch flows through the electromagnetic valve 12 and enters the medicine spraying pipeline; each spraying pipeline is divided into three pipelines, liquid medicine from different pesticide boxes is respectively received, and when the weeding operation vehicle moves forwards, each system is started to work, so that different types of herbicides are sprayed according to different weed types, and a basic spraying function is completed.
The image acquisition system 9 sends the acquired weed images to the weed identification and classification system 8, the weed identification and classification system 8 analyzes and processes the acquired weed images, then sends the identified weed types to the spraying control and decision system 7, and finally the spraying control and decision system 7 sprays herbicide.
The workflow of the weed identification and classification system 8 is:
s1: the method comprises the steps of collecting a data set, wherein the height of a shot image is about 30-50 cm when weeds are collected in the early stage, and the angle is about 60-90 degrees, namely, the shooting angle is close to the vertical weeds, and the recognition accuracy of the follow-up weeds can be influenced due to different weather, illumination and other conditions, so that the weed images are collected on sunny days and cloudy days respectively;
s2: labeling the pictures into an xml file in a VOC format by using labeling software LabelImg, labeling weeds by using a rectangular frame, wherein the frame displays the name and position information of the picture labeling, and before the pictures are input into a convolution network to extract information, carrying out image normalization processing, image normalization processing and image feature extraction on the acquired pictures;
s21: carrying out image normalization; the range of the pixel points of the acquired RGB image is between 0 and 255, and in order to accelerate network training, the pixel is converted into between 0 and 1 by using a formula (1) to perform normalization operation;
wherein: x is X i Pixel value, X, representing the i-th point in the image max And x min Representing the maximum and minimum values of the image pixels, respectively;
s22: performing image standardization processing; because the data information of the image is distributed more dispersedly, the difficulty of network training is increased, and the image standardization is needed, the main principle is that the data is subjected to centering treatment through mean removal, the mean value and variance of the pixel points are obtained through formulas (2) and (3), and then the standardization treatment is carried out through formula (4);
wherein: μ is the mean of the pixels of the image, σ represents the standard deviation of the pixels, N represents the number of pixels of the image X;
s23: extracting image features; in the process of feature extraction, low-level features such as textures, frames and colors are generally extracted by a shallow network, high-level and abstract features are extracted by a deep network, and the main purpose is to fit a picture target with a marked real area so as to achieve the minimum error rate; extracting the characteristic information of the neural network from the processed picture through a formula (5);
wherein: z (u, v) represents the pixel value of the image in the v column of the u th row corresponding to the characteristic image after convolution, and x i，j Representing the pixel value of the image of the ith row and the jth column corresponding to the convolved image, and k represents the convolution kernel;
s3: inputting the picture into a convolution network to extract information; the size of the image after convolution operation is changed, mainly the number of channels is increased and the length and width are changed; the neural network training process is a process of continuously updating parameter fitting training data back and forth, and in order to improve the network optimization speed, the feature map and the performance are subjected to standardized processing, so that the convergence capacity of the network is enhanced;
after the network training is finished, a better weight value can be obtained among all layers, the optimal prediction effect can be achieved, and the weight is reserved for detecting weeds in the follow-up process; based on the previously obtained network weight, the picture obtained in the actual environment is processed and then is transmitted into the network, because the weight parameters among layers reach a better level in the training process, the input image only needs to be transmitted forward under the parameters, and the actual target class probability is finally obtained;
wherein: l represents the loss function, p, of the model i A set of data, t, representing model predictions i Representing data corresponding to a real sample of a model, N cls The predicted sample size, i, represents the ith bit data value, L in the array cls Model classification loss function, λ represents weight, N reg Number of samples regressed, L reg Model regression loss function,Representing the category of the prediction->Representing the true category;
adopting precision P, recall rate R and average precision mean value m AP As the evaluation index of target detection, the three values are all in the range of 0,1]At the same time introduce F 1 Carrying out harmonic mean evaluation on the values;
the precision represents the ratio of the number of a certain type of weeds detected correctly to the number predicted as the type of weeds, and the recall rate represents the ratio of all targets of the type of weeds in the sample to be predicted correctly;
wherein: t (T) P Representing the number of correctly divided positive samples;
F P indicating the number of erroneous divisions into positive samples;
T N representing the number of correctly divided negative samples;
F N indicating the number of false divisions into negative samples.
The average precision represents the detection effect of the used detection network model on a certain class of targets, the larger the value of the average precision represents the better overall detection effect, the worse the overall detection effect is, the average precision is mainly represented by a precision and recall curve, namely a PR curve, the abscissa of the curve is the recall and reflects the covering capability of an aligned sample, the precision of a vertical axis of the curve reflects the precision of a predicted positive sample, and the calculation of the average precision is taken as the integral of the precision and recall curve on [0,1 ];
the average precision mean value represents the mean value of the average precision of all the categories in the data set, and the calculation method is the ratio of the sum of the average precision of all the categories to the number of all the categories;
wherein: a is that p Representing average precision, R representing recall, m AP Represents the average precision mean value, n represents the category number of all weeds;
F 1 the value is a comprehensive evaluation index based on precision and recall, which is defined as a harmonic mean of precision and recall;
the weed identification result is displayed on the computer end in real time, and the identified weed type and the identified probability are fed back according to the performance of the pre-training network structure.
The spraying control and decision system 7 judges the existence of weeds and the types of weeds according to different types of weed information fed back by the weed identification and classification system 8, sends corresponding control instructions according to the identification result, and opens the corresponding electromagnetic valve 12 to spray the corresponding types of herbicides.
The image acquisition system 9 is an industrial vision camera.
The liquid level sensor 5 is fixedly connected in the medicine chest 3, and the liquid level sensor 5 is electrically connected with the medicine spraying control and decision system 7.
The medicine chest 3 is provided with a medicine inlet 4.
As shown in fig. 6 to 7, the spray heads are placed in parallel with the forward direction of the working vehicle, 3 rows of spray heads are installed at adjacent 20cm positions, respectively, and 3 spray heads 1 for receiving the liquid medicine from three different liquid medicine tanks 3 are placed at each position, and each spray head is individually controlled by one solenoid valve 12, in order to improve the working efficiency of the spraying working vehicle.
The image collecting device adopted by the image collecting system 9 is an industrial vision camera, and can realize inclination angle correction and image distortion correction; in addition, the method has the characteristics of higher recognition precision (the error of the positioning precision is less than 2 mm), short single recognition time (the time for photographing, processing and data output is less than 1 second), and positioning and classifying of multiple targets.
The weed identification and classification system 8 identifies weeds through convolutional neural network models (Convolutional Neural Network, CNN), the selected network model is a Yolov5 model in YOLO (You Only Look Once) series models based on convolutional neural network single-stage detection, and a threshold value is set to IoU =0.5, and the number of training pictures per batch is bacterial size=4. The hardware system for training the network model of the system is a desktop workstation, the processor is AMD Ryzen threadripper 3970x 32-core processor64, the running memory is 32GB, and the Graphic Processing Unit (GPU) is NVIDIA Corporation GeForce RTX3060ti. The running system is Ubuntu20.4; selecting a Pytorch deep learning framework supporting a plurality of neural network algorithms;
the weed identification and classification system 8 will be pre-trained prior to field operation to achieve optimal performance and parameters for the identification model. Since pretraining requires a large number of weed data sets, and the types of weeds in the field are 77 families and comprise 580 types, the weeds are roughly classified into 3 categories of grasses, broadleaf weeds and sedge according to main morphological characteristics, and 3 categories of labels are prepared according to grasses, broadleaf weeds and sedge weeds when the weed data sets are prepared;
the above disclosure is merely illustrative of specific embodiments of the present invention, but the present invention is not limited thereto, and any variations that can be considered by those skilled in the art should fall within the scope of the present invention.
Claims (10)
1. Accurate medicine weeding device that spouts based on weeds kind, its characterized in that includes: the system comprises an image acquisition system (9), a weed identification and classification system (8), a pesticide spraying control and decision system (7), a target spraying weeding system and a trolley travelling system (6);
the system comprises an image acquisition system (9), a weed identification and classification system (8), a pesticide spraying control and decision system (7) and a weed identification and classification system (8), wherein the image acquisition system (9) is fixed on a U-shaped frame at the front end of the trolley traveling system (6), and the weed identification and classification system (8) is fixed at the front end of a chassis of the trolley traveling system (6) and is fixedly connected with the trolley traveling system (6) at the rear side of the weed identification and classification system (8);
the image acquisition system (9) is electrically connected with the weed identification and classification system (8), and the weed identification and classification system (8) and the target spraying weeding system are electrically connected with the spraying control and decision system (7) respectively.
2. The weed species-based precision spray weeding device according to claim 1, wherein: the targeted spraying weeding system comprises a group of medicine boxes (3), wherein the medicine boxes (3) are arranged side by side at the rear end of a chassis of the trolley traveling system (6), the medicine boxes (3) are fixedly connected with the chassis of the trolley traveling system (6), the medicine boxes (3) are provided with medicine outlets (2), the medicine boxes (3) correspond to one ends of a loop pipe fixedly connected with the medicine boxes (3), the other ends of the loop pipe are fixedly connected with the medicine boxes (3), a filter (10), a medicine pump (11), electromagnetic valves (12) and safety valves (13) are arranged on the loop pipe, each electromagnetic valve (12) is fixedly connected with a group of spray heads (1), and each spray head (1) is respectively connected with the corresponding electromagnetic valve (12).
3. The weed species-based precision spray weeding device according to claim 2, wherein: the system is characterized in that the image acquisition system (9) sends acquired weed images to the weed identification and classification system (8), the weed identification and classification system (8) analyzes and processes the acquired weed images, then sends identified weed types to the pesticide spraying control and decision system (7), and finally the pesticide spraying control and decision system (7) sprays herbicide.
4. The weed species-based precision spray weeding device according to claim 2, wherein: the workflow of the weed identification and classification system (8) is as follows:
s1: the method comprises the steps of collecting a data set, wherein the height of a shot image is about 30-50 cm when weeds are collected in the early stage, and the angle is about 60-90 degrees, namely, the shooting angle is close to the vertical weeds, and the recognition accuracy of the follow-up weeds can be influenced due to different weather, illumination and other conditions, so that the weed images are collected on sunny days and cloudy days respectively;
s2: labeling the pictures into an xml file in a VOC format by using labeling software LabelImg, labeling weeds by using a rectangular frame, wherein the frame displays the name and position information of the picture labeling, and before the pictures are input into a convolution network to extract information, carrying out image normalization processing, image normalization processing and image feature extraction on the acquired pictures;
s21: carrying out image normalization; the range of the pixel points of the acquired RGB image is between 0 and 255, and in order to accelerate network training, the pixel is converted into between 0 and 1 by using a formula (1) to perform normalization operation;
wherein: pixel point values representing a first point in the image, and maximum and minimum values representing pixels of the image, respectively;
s22: performing image standardization processing; because the data information of the image is distributed more dispersedly, the difficulty of network training is increased, and the image standardization is needed, the main principle is that the data is subjected to centering treatment through mean removal, the mean value and variance of the pixel points are obtained through formulas (2) and (3), and then the standardization treatment is carried out through formula (4);
wherein: the mean value of the pixel points of the image, the standard deviation of the pixel points, and N represents the pixel quantity of the image;
s23: extracting image features; in the process of feature extraction, low-level features such as textures, frames and colors are generally extracted by a shallow network, high-level and abstract features are extracted by a deep network, and the main purpose is to fit a picture target with a marked real area so as to achieve the minimum error rate; extracting the characteristic information of the neural network from the processed picture through a formula (5);
wherein: z (u, v) represents the pixel value of the image in the v column of the u th row corresponding to the characteristic image after convolution, and x i，j Representing the pixel value of the image of the ith row and the jth column corresponding to the convolved image, and k represents the convolution kernel;
s3: inputting the picture into a convolution network to extract information; the size of the image after convolution operation is changed, mainly the number of channels is increased and the length and width are changed; the neural network training process is a process of continuously updating parameter fitting training data back and forth, and in order to improve the network optimization speed, the feature map and the performance are subjected to standardized processing, so that the convergence capacity of the network is enhanced;
after the network training is finished, a better weight value can be obtained among all layers, the optimal prediction effect can be achieved, and the weight is reserved for detecting weeds in the follow-up process; based on the previously obtained network weight, the picture obtained in the actual environment is processed and then is transmitted into the network, because the weight parameters among layers reach a better level in the training process, the input image only needs to be transmitted forward under the parameters, and the actual target class probability is finally obtained;
wherein: l represents the loss function, p, of the model i A set of data, t, representing model predictions i Representing data corresponding to a real sample of a model, N cls The predicted sample size, i, represents the ith bit data value, L in the array cls Model classification loss function, λ represents weight, N reg Number of samples regressed, L reg Model regression loss function,Representing the category of the prediction->Representing the true category.
5. The weed species based precision spray weeding device according to claim 4, wherein: adopting precision P, recall rate R and average precision mean value m AP As the evaluation index of target detection, the three values are all in the range of 0,1]At the same time introduce F 1 Carrying out harmonic mean evaluation on the values;
the precision represents the ratio of the number of a certain type of weeds detected correctly to the number predicted as the type of weeds, and the recall rate represents the ratio of all targets of the type of weeds in the sample to be predicted correctly;
wherein: t (T) P Representing the number of correctly divided positive samples;
F P indicating the number of erroneous divisions into positive samples;
T N representing the number of correctly divided negative samples;
F N indicating the number of false divisions into negative samples.
6. The weed species based precision spray weeding device according to claim 5, wherein: the average precision represents the detection effect of the used detection network model on a certain class of targets, the larger the value of the average precision represents the better overall detection effect, the worse the overall detection effect is, the average precision is mainly represented by a precision and recall curve, namely a PR curve, the abscissa of the curve is the recall and reflects the covering capability of an aligned sample, the precision of a vertical axis of the curve reflects the precision of a predicted positive sample, and the calculation of the average precision is taken as the integral of the precision and recall curve on [0,1 ];
A P ＝∫ 0 P(R)dR (9)
the average precision mean value represents the mean value of the average precision of all the categories in the data set, and the calculation method is the ratio of the sum of the average precision of all the categories to the number of all the categories;
wherein: a is that p Representing average precision, R representing recall, m AP Represents the average precision mean value, n represents the category number of all weeds;
F 1 the value is a comprehensive evaluation index based on precision and recall, which is defined as a harmonic mean of precision and recall;
the weed identification result is displayed on the computer end in real time, and the identified weed type and the identified probability are fed back according to the performance of the pre-training network structure.
7. The weed species-based precision spray weeding device according to claim 1, wherein: the spraying control and decision system 7 judges the existence of weeds and the types of weeds according to different types of weed information fed back by the weed identification and classification system 8, sends corresponding control instructions according to the identification result, and opens corresponding electromagnetic valves (12) to spray the corresponding types of herbicides.
8. The weed species based precision spray weeding device according to claim 7, wherein: the image acquisition system (9) is an industrial vision camera.
9. The weed species-based precision spray weeding device according to claim 2, wherein: the liquid level sensor (5) is fixedly connected in the medicine chest (3), and the liquid level sensor (5) is electrically connected with the medicine spraying control and decision system (7).
10. The weed species-based precision spray weeding device according to claim 2, wherein: the medicine box (3) is provided with a medicine inlet (4).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310818917.1A CN116868974A (en) | 2023-07-05 | 2023-07-05 | Accurate medicine weeding device that spouts based on weeds kind |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310818917.1A CN116868974A (en) | 2023-07-05 | 2023-07-05 | Accurate medicine weeding device that spouts based on weeds kind |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116868974A true CN116868974A (en) | 2023-10-13 |
Family
ID=88269011
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310818917.1A Pending CN116868974A (en) | 2023-07-05 | 2023-07-05 | Accurate medicine weeding device that spouts based on weeds kind |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116868974A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117292248A (en) * | 2023-10-30 | 2023-12-26 | 吉林农业大学 | Deep learning-based farmland pesticide spraying system and weed detection method |
| CN117882697A (en) * | 2024-03-14 | 2024-04-16 | 东海实验室 | Accurate and rapid intelligent laser weeding device and method |
-
2023
- 2023-07-05 CN CN202310818917.1A patent/CN116868974A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117292248A (en) * | 2023-10-30 | 2023-12-26 | 吉林农业大学 | Deep learning-based farmland pesticide spraying system and weed detection method |
| CN117292248B (en) * | 2023-10-30 | 2024-04-26 | 吉林农业大学 | Deep learning-based farmland pesticide spraying system and weed detection method |
| CN117882697A (en) * | 2024-03-14 | 2024-04-16 | 东海实验室 | Accurate and rapid intelligent laser weeding device and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116868974A (en) | Accurate medicine weeding device that spouts based on weeds kind | |
| CN105173085B (en) | Unmanned plane variable farm chemical applying automatic control system and method | |
| Tian | Development of a sensor-based precision herbicide application system | |
| US11375655B2 (en) | System and method for dispensing agricultural products into a field using an agricultural machine based on cover crop density | |
| CN113439727B (en) | Deinsectization method, device, equipment and storage medium for greenhouse crops | |
| CN108549869A (en) | A kind of adaptive operational method of plant protection drone based on expert system | |
| CN111596689A (en) | Intelligent agricultural plant protection operation control system based on big data Internet of things | |
| CN102564593A (en) | Plant growth condition monitoring system based on compute vision and internet of things | |
| CN102172233A (en) | Method for carrying out real-time identification and targeted spraying on cotton field weeds | |
| CN102428904A (en) | Automatic targeting and variable atomizing flow control system for weeding robot | |
| WO2021226900A1 (en) | Cotton crop row detection method and apparatus based on computer vision, and storage medium | |
| CN113158750A (en) | Self-feedback learning evaluation method of plant growth model based on convolutional neural network | |
| Vikram | Agricultural Robot–A pesticide spraying device | |
| CN109003198A (en) | A kind of precision agriculture management platform and method based on big data technology | |
| CN113142170B (en) | Unmanned aerial vehicle intelligent fixed-point weeding technology suitable for rice field | |
| CN113645844A (en) | Method for treating plants in a field of plants with variable application rates | |
| CN114514914A (en) | Intelligent sensing fertilization and pesticide spraying method and device | |
| Zhang et al. | Variable rate air‐assisted spray based on real‐time disease spot identification | |
| CN116818002A (en) | Intelligent agricultural planting monitoring system based on big data | |
| CN112712088B (en) | Animal fat condition detection method and device and computer readable storage medium | |
| Hong et al. | Adaptive target spray system based on machine vision for plant protection UAV | |
| Mary et al. | IoT Based Weed Detection and Removal in Precision Agriculture | |
| CN115187794A (en) | Data detection and analysis system for landscape plant growth environment | |
| Chang et al. | Straight-line generation approach using deep learning for mobile robot guidance in lettuce fields | |
| CN112385632A (en) | Variable spraying control system and control method of plant protection unmanned aerial vehicle based on LQR (Long Range response) controller |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |