Pesticide delivery method, device, equipment and storage medium

Abstract

The present invention relates to the field of pesticide delivery technology, and discloses a pesticide delivery method, device, equipment, and storage medium. The method comprises: real-time acquisition of grape leaf images in a vineyard, detection of the grape leaf images using a grape pest and disease Yolov5 detection model to obtain pest and disease data; executing a preset pesticide spraying strategy based on the pest and disease data and recording reference data; gradually adjusting the preset pesticide spraying strategy based on the reference data to obtain a target pesticide spraying strategy; and establishing a pesticide delivery model based on the target pesticide spraying strategy and the reference data to achieve automatic pesticide delivery. The present invention uses the detection model to detect grape leaf images to obtain pest and disease data, thereby determining the target pesticide spraying strategy and establishing a pesticide delivery model to achieve automatic pesticide delivery. This solves the problem that traditional methods require a lot of time and effort and have low accuracy, achieves automation, improves efficiency, and achieves high delivery accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of pesticide delivery, and in particular to a pesticide delivery method, device, equipment and storage medium. Background Art

[0002] There are many types of grape leaf diseases, and the diseases are highly similar. Using traditional HOG feature extraction or CNN network training methods requires a lot of time and energy and has low accuracy, which cannot be well combined with the actual situation. Summary of the Invention

[0003] The main purpose of the present invention is to provide a pesticide delivery method, device, equipment and storage medium, aiming to solve the technical problems that the existing technology requires a lot of time and effort, has low accuracy, and cannot be well integrated with actual conditions.

[0004] To achieve the above object, the present invention provides a method for applying pesticides, comprising the following steps:

[0005] Real-time collection of grape leaf images in the vineyard, detection of the grape leaf images using the grape pest and disease Yolov5 detection model, and obtaining pest and disease data, wherein the pest and disease data includes the number of leaves in the vineyard with grape pests and diseases and the types of pests and diseases to which different leaves belong;

[0006] Executing a preset pesticide spraying strategy based on the pest and disease data and recording reference data, wherein the reference data includes the time of pesticide spraying, the recovery of grape leaf pests and diseases in the sprayed area, and the pest and disease characteristics of each grape leaf treated with the pesticide;

[0007] Stepwise adjusting the preset pesticide spraying strategy according to the reference data to obtain a target pesticide spraying strategy, wherein the pesticide spraying strategy includes the concentration, type and interval of the spraying pesticide;

[0008] A pesticide delivery model is established based on the target pesticide spraying strategy and reference data to achieve automatic pesticide delivery.

[0009] Optionally, after establishing a pesticide delivery model based on the target pesticide spraying strategy and reference data to achieve automatic pesticide delivery, the method further includes:

[0010] Storing the pest and disease data and the reference data as a plurality of experience samples, and calculating the accuracy and recall of the data in each experience sample;

[0011] The pesticide delivery model is trained according to the accuracy and recall of the data in each experience sample.

[0012] Optionally, the step of gradually adjusting the preset pesticide spraying strategy according to the reference data to obtain a target pesticide spraying strategy includes:

[0013] Increase or decrease the concentration of sprayed pesticides according to the recovery of grape leaf diseases and pests in the sprayed area until the optimal concentration of sprayed pesticides is determined;

[0014] Increase or decrease the interval between spraying pesticides according to the recovery of grape leaf pests and diseases in the sprayed area until the optimal spraying interval is determined;

[0015] Select several types of spraying pesticides for alternating use based on the recovery of grape leaf diseases and insect pests in the sprayed area;

[0016] The optimal pesticide spraying concentration, the optimal pesticide spraying time interval, and the types of several pesticides to be sprayed are used as target pesticide spraying strategies.

[0017] Optionally, before collecting grape leaf images in the vineyard in real time and detecting the grape leaf images using the grape pest and disease Yolov5 detection model to obtain pest and disease data, the method further includes:

[0018] Obtaining leaf images of grape diseases and insect pests, and performing data amplification on the leaf images of grape diseases and insect pests to obtain a data set;

[0019] Dividing the data set into a training set and a validation set according to a preset ratio;

[0020] Constructing a Yolov5 detection model for grape pests and diseases, and adding an adaptive attention module and a feature enhancement module to the feature pyramid structure of the Yolov5 detection model to obtain an improved Yolov5 detection model for grape pests and diseases, wherein the Yolov5 detection model for grape pests and diseases includes an input layer, a backbone layer, a neck layer, and a prediction layer;

[0021] Improve the Yolov5 grape pest and disease detection model through iterative training of the training set until convergence and obtain the optimal network weights;

[0022] The optimal network weight is input into the improved grape disease and insect pest Yolov5 detection model, and the improved grape disease and insect pest Yolov5 detection model is optimized to obtain an optimized improved grape disease and insect pest Yolov5 detection model.

[0023] Optionally, the iterative training of the training set to improve the Yolov5 detection model for grape pests and diseases until convergence to obtain the optimal network weights includes:

[0024] Inputting the training set into the backbone layer through the input layer for feature extraction to obtain a feature map of the training set;

[0025] Performing a top-down downsampling operation and a fusion operation on the feature map of the training set through the neck layer to obtain a feature map of the training set;

[0026] Perform prediction in the prediction layer based on the feature map of the training set, calculate the loss, and update the model parameters;

[0027] Retrain until the loss function converges or the maximum number of iterations is reached to obtain the optimal network weights.

[0028] Optionally, the feature maps of the training set include a first feature map, a second feature map, a third feature map, a fourth feature map, and a fifth feature map;

[0029] The feature graphs of the training set include a first feature graph, a second feature graph, and a third feature graph;

[0030] The step of performing a top-down downsampling operation and a fusion operation on the feature map of the training set through the neck layer to obtain a feature map of the training set includes:

[0031] The fifth feature map is generated by the adaptive attention module into a sixth feature map, and the sixth feature map is fused with the fifth feature map and input into the feature enhancement module for feature enhancement to obtain a first feature map;

[0032] Downsampling the first feature map to generate a seventh feature map, fusing the seventh feature map with the fourth feature map and inputting the seventh feature map into the feature enhancement module for feature enhancement to obtain a second feature map;

[0033] The second feature map is down-sampled to generate an eighth feature map, which is fused with the third feature map and then input into the feature enhancement module for feature enhancement to obtain a third feature map.

[0034] Optionally, generating a sixth feature map from the fifth feature map through the adaptive attention module includes:

[0035] Obtaining a preset number of semantic features through an adaptive pooling layer in the adaptive attention module according to the fifth feature map, wherein each of the preset number of semantic features has a different scale;

[0036] Performing a convolution operation on the preset number of semantic features to convert the preset number of semantic features into the same channel dimension;

[0037] Performing an upsampling operation on the preset number of semantic features, and fusing the preset number of semantic features through a Concat layer to obtain a target feature map;

[0038] Obtaining a spatial weight of the target feature map;

[0039] Separating the spatial weight and the target feature map after performing a Hadamard product operation to obtain a preset number of new semantic features;

[0040] A matrix addition operation is performed on the preset number of new semantic features and the fifth feature map to obtain a sixth feature map.

[0041] In addition, to achieve the above-mentioned purpose, the present invention further provides a pesticide delivery device, which includes:

[0042] In addition, to achieve the above-mentioned purpose, the present invention also proposes a pesticide delivery device, which includes: a memory, a processor, and a pesticide delivery program stored on the memory and executable on the processor, wherein the pesticide delivery program is configured to implement the steps of the pesticide delivery method described above.

[0043] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which a pesticide delivery program is stored. When the pesticide delivery program is executed by a processor, the steps of the pesticide delivery method described above are implemented.

[0044] The present invention collects grape leaf images from a vineyard in real time, detects the grape leaf images using the grape pest and disease Yolov5 detection model, and obtains pest and disease data. A preset pesticide spraying strategy is executed based on the pest and disease data, and reference data is recorded. The preset pesticide spraying strategy is gradually adjusted based on the reference data to obtain a target pesticide spraying strategy. A pesticide delivery model is established based on the target pesticide spraying strategy and the reference data to achieve automatic pesticide delivery. This method, by detecting grape leaf images using a detection model to obtain pest and disease data, determines the target pesticide spraying strategy, and establishes a pesticide delivery model to achieve automatic pesticide delivery. This addresses the time and effort required and low accuracy of traditional methods, achieving automation, improving efficiency, and achieving high delivery accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a structural diagram of a pesticide delivery device in a hardware operating environment according to an embodiment of the present invention;

[0046] Figure 2 This is a schematic flow chart of a first embodiment of the pesticide delivery method of the present invention;

[0047] Figure 3This is a schematic diagram of the overall process of an embodiment of the pesticide delivery method of the present invention;

[0048] Figure 4 This is a schematic flow chart of a second embodiment of the pesticide delivery method of the present invention;

[0049] Figure 5 This is a schematic flow chart of a third embodiment of the pesticide delivery method of the present invention;

[0050] Figure 6 This is a schematic diagram of the structure of the feature enhancement module in the third embodiment of the pesticide delivery method of the present invention;

[0051] Figure 7 Schematic diagram of the improved characteristic pyramid structure in the third embodiment of the pesticide delivery method of the present invention;

[0052] Figure 8 This is a schematic diagram of the structure of the adaptive attention module in the third embodiment of the pesticide delivery method of the present invention;

[0053] Figure 9 This is a structural block diagram of the first embodiment of the pesticide delivery device of the present invention.

[0054] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0055] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0056] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of the pesticide delivery device in the hardware operating environment involved in the embodiment of the present invention.

[0057] like Figure 1As shown, the pesticide delivery device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0058] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the pesticide delivery device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0059] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a pesticide delivery program.

[0060] exist Figure 1 In the pesticide delivery device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the pesticide delivery device of the present invention can be set in the pesticide delivery device, and the pesticide delivery device calls the pesticide delivery program stored in the memory 1005 through the processor 1001 and executes the pesticide delivery method provided by the embodiment of the present invention.

[0061] The embodiment of the present invention provides a method for applying pesticides, referring to Figure 2 , Figure 2 This is a schematic flow chart of the first embodiment of the pesticide delivery method of the present invention.

[0062] In this embodiment, the pesticide delivery method includes the following steps:

[0063] Step S10: Real-time collection of grape leaf images in the vineyard, detection of the grape leaf images using the grape pest and disease Yolov5 detection model, and obtaining pest and disease data, wherein the pest and disease data includes the number of leaves in the vineyard with grape pests and diseases and the types of pests and diseases to which different leaves belong.

[0064] It should be noted that the pests and diseases described include grape black rot, grapevine eska virus, and grapevine leaf blight, each characterized by different disease characteristics. Grape black rot is characterized by the appearance of reddish-brown lesions. As the disease progresses, the lesions expand and deepen, forming black, raised spots with dark-brown edges, an off-white center, and brown outer edges. Grapevine eska virus is characterized by necrosis of leaf veins, impeding nutrient transport, and in severe cases, leaf loss and branch withering from the tips. Grape leaf blight is characterized by a bluish-green or dark green edge in the early stages of infection, with a cloud-like appearance. After one to two days, the affected area turns yellowish-brown, and the cloud-like, bluish-green lesions spread inward, forming a circular ring. In the later stages of infection, the entire leaf will wilt.

[0065] It can be understood that the images of grape leaves in the vineyard are collected in real time by a camera device, and are detected using the pre-built and trained grape pest and disease detection model Yolov5 to determine the number of leaves in the vineyard with grape pests and diseases and the types of pests and diseases to which different leaves belong.

[0066] In the specific implementation, dense sampling is performed evenly at different locations in the image. Different aspect ratios can be used during sampling. Then, the grape pest and disease detection model Yolov5 is used to extract features and then directly perform classification and regression to determine whether a leaf has a certain type of pest and disease, and record the pest and disease characteristics of each grape leaf.

[0067] Step S20: executing a preset pesticide spraying strategy according to the pest and disease data and recording reference data, wherein the reference data includes the time of pesticide spraying, the recovery of pests and diseases of grape leaves in the sprayed area, and the pest and disease characteristics of each grape leaf treated with pesticides.

[0068] It should be noted that the pesticide spraying strategy includes the concentration, type and interval of the spraying pesticide. For example, the preset pesticide spraying strategy for grape black rot is to spray 800-1000 times of 50% thiophanate-methyl wettable powder every three days.

[0069] It can be understood that the corresponding preset pesticide spraying strategy is executed according to the number of leaves with grape pests and diseases in the vineyard and the types of pests and diseases to which different leaves belong, and reference data such as the time of pesticide spraying, the recovery of grape leaf pests and diseases in the sprayed area, and the pest and disease characteristics of each grape leaf treated with pesticides are recorded.

[0070] Step S30: gradually adjusting the preset pesticide spraying strategy according to the reference data to obtain a target pesticide spraying strategy, wherein the pesticide spraying strategy includes the concentration, type and interval time of the sprayed pesticide.

[0071] It is understandable that the concentration, type and interval of spraying pesticides should be gradually adjusted according to the recovery of grape leaf diseases and pests in the spraying area until the concentration, type and interval with the best treatment effect are determined.

[0072] Furthermore, in order to determine the target strategy, step S30 includes: increasing or decreasing the concentration of the sprayed pesticide according to the recovery of grape leaf diseases and insect pests in the sprayed area until the optimal spraying concentration is determined; increasing or shortening the time interval for spraying the pesticide according to the recovery of grape leaf diseases and insect pests in the sprayed area until the optimal spraying time interval is determined; selecting several types of spraying pesticides for alternating use according to the recovery of grape leaf diseases and insect pests in the sprayed area; and using the optimal spraying concentration of the pesticide, the optimal spraying time interval of the pesticide, and the several types of spraying pesticides as the target pesticide spraying strategy.

[0073] It is understandable that if the concentration of the sprayed pesticide is insufficient, the recovery of grape leaf diseases and pests in the sprayed area may be slow. At this time, the concentration of the sprayed pesticide needs to be increased. If the concentration of the sprayed pesticide is too high, it may cause damage to the grape leaves in the sprayed area and aggravate the disease situation. At this time, the concentration of the sprayed pesticide needs to be reduced.

[0074] It is understandable that if the time interval between spraying pesticides is too long, it may not be effective for the grape leaf diseases and insect pests in the sprayed area. In this case, the time interval between spraying pesticides needs to be shortened. If the time interval between spraying pesticides is too short, it may cause waste for the grape leaf diseases and insect pests in the sprayed area under the same recovery conditions. In this case, the time interval between spraying pesticides needs to be increased.

[0075] It is understandable that several types of spraying pesticides with the same effect are selected and used alternately according to the recovery of grape leaf diseases and insect pests in the sprayed area to prevent the development of pesticide resistance.

[0076] Step S40: establishing a pesticide delivery model according to the target pesticide spraying strategy and reference data to achieve automatic pesticide delivery.

[0077] It can be understood that the pesticide delivery model constructed based on the Yolov5 detection model for grape pests and diseases takes the changes in pest and disease characteristics, the type and concentration of pesticides applied, and the length of time after pesticide application as inputs. It automatically selects corresponding spraying strategies for different types of plant pests and diseases to achieve automatic pesticide delivery and automation, thereby maximizing both fruit yield and efficiency in the orchard.

[0078] like Figure 3 As shown, Figure 3 This is a schematic diagram of the overall process of the pesticide application method of this embodiment. In the figure, various grape leaves are recorded, and the types of pests and diseases described on different leaves are recorded. The different types of leaves are divided into training sets and validation sets at a ratio of 9:1. A Yolov network model for grape pests and diseases is constructed. The Yolov network model for grape pests and diseases is specifically a Yolov5 network model for grape pests and diseases. The Yolov network model data for grape pests and diseases is adjusted according to the characteristics of the pest and disease leaves. The Yolov network model for grape pests and diseases is specifically a Yolov5 network model for grape pests and diseases. The required type of pesticide is obtained based on the type and amount of pests and diseases on different plants, and the right medicine is prescribed to solve the grape pest and disease problem.

[0079] This embodiment collects grape leaf images from a vineyard in real time, detects the grape leaf images using the grape pest and disease Yolov5 detection model, and obtains pest and disease data. A preset pesticide spraying strategy is executed based on the pest and disease data and reference data is recorded. The preset pesticide spraying strategy is gradually adjusted based on the reference data to obtain a target pesticide spraying strategy. A pesticide delivery model is established based on the target pesticide spraying strategy and the reference data to achieve automatic pesticide delivery. Through the above method, pest and disease data is obtained by detecting grape leaf images using a detection model, thereby determining a target pesticide spraying strategy and establishing a pesticide delivery model to achieve automatic pesticide delivery. This solves the problem that traditional methods require a lot of time and effort and have low accuracy, achieving automation, improving efficiency, and achieving high delivery accuracy.

[0080] refer to Figure 4 , Figure 4 This is a flow chart of the second embodiment of the pesticide delivery method of the present invention.

[0081] Based on the first embodiment described above, the pesticide application method of this embodiment further includes, before step S40:

[0082] Step S41: storing the pest and disease data and the reference data as a plurality of experience samples, and calculating the accuracy and recall rate of the data in each experience sample.

[0083] It should be noted that each of the plurality of experience samples records the pest and disease characteristics of the training set.

[0084] It can be understood that the accuracy is the proportion of all correct predictions (positive and negative) to the total, the recall rate is the proportion of correctly predicted positive samples to all actual positive samples, and the recall rate is the probability of being predicted as a positive sample among the actually positive samples. A high recall rate means that there may be more false detections.

[0085] In the specific implementation, the accuracy calculation formula is:

[0086] Precision = TP / (TP+FP)

[0087] Among them, Precision is the accuracy rate, TP is the area where the pesticide delivery model is delivered correctly, and FP is the area where the pesticide delivery model is delivered correctly.

[0088] The recall calculation formula is:

[0089] Recall = TP / (TP+FN)

[0090] Among them, Recall is the recall rate, TP is the area where the pesticide delivery model is correctly delivered, and FN is the area where the disease exists but the pesticide delivery model has not delivered it.

[0091] Step S42: training the pesticide delivery model according to the accuracy and recall of the data in each experience sample.

[0092] It can be understood that each experience sample records the pest and disease characteristics of the training set, and the pesticide delivery model is trained according to the accuracy and recall rate corresponding to each experience sample, and the parameters of the pesticide delivery model are updated to improve the accuracy of the model delivery.

[0093] This embodiment stores the pest and disease data and the reference data as several empirical samples, calculates the accuracy and recall of the data in each empirical sample, and trains the pesticide delivery model based on the accuracy and recall of the data in each empirical sample. Through this method, the pesticide delivery model is trained based on the accuracy and recall corresponding to each empirical sample, updating the parameters of the pesticide delivery model and improving the accuracy of the model delivery, thereby maximizing both fruit yield and efficiency in the orchard.

[0094] refer to Figure 5 , Figure 5 Schematic diagram of the flow chart of the third embodiment of the pesticide delivery method of the present invention.

[0095] Based on the first embodiment described above, the pesticide application method of this embodiment further includes, before step S10:

[0096] Step S01: obtaining leaf images of grape diseases and insect pests, and performing data amplification on the leaf images of grape diseases and insect pests to obtain a data set.

[0097] It should be noted that, since machine learning requires a large number of image samples for learning, the leaf images of the grape pests and diseases are amplified to obtain more images to form a data set.

[0098] It is understandable that a preset number of grape leaf images of pests and diseases are obtained, and the preset number of grape leaf images of pests and diseases are randomly cropped, scaled, and randomly arranged and spliced ​​to perform data enhancement to obtain new grape leaf images of pests and diseases. The new grape leaf images of pests and diseases are added to the preset number of grape leaf images of pests and diseases to achieve data amplification, and the grape leaf images of pests and diseases after data amplification are used as a data set. The preset number can be 5000, 6000, 7000, etc., and this embodiment does not impose a specific limitation on this.

[0099] Step S02: Divide the data set into a training set and a validation set according to a preset ratio.

[0100] It should be noted that the preset ratio may be 9:1, that is, the dataset of wine cap packaging defects is divided into a training set and a validation set at a ratio of 9:1.

[0101] Step S03: constructing a Yolov5 detection model for grape pests and diseases, and adding an adaptive attention module and a feature enhancement module to the feature pyramid structure of the Yolov5 detection model for grape pests and diseases to obtain an improved Yolov5 detection model for grape pests and diseases, wherein the Yolov5 detection model for grape pests and diseases includes an input layer, a backbone layer, a neck layer, and a prediction layer.

[0102] It should be noted that a Yolov5 detection model for grape pests and diseases is constructed, and the traditional feature pyramid structure (FPN) in the Yolov5 detection model for grape pests and diseases is replaced with an improved feature pyramid structure (AF-FPN), wherein the improved feature pyramid structure (AF-FPN) is obtained by adding an adaptive attention module and a feature enhancement module to the traditional feature pyramid structure (FPN).

[0103] It can be understood that the adaptive attention module reduces the loss of contextual information in high-level feature maps due to the reduction of feature channels, and the feature enhancement module enhances the representation of the feature pyramid and speeds up the reasoning.

[0104] It is worth noting that the grape pest and disease Yolov5 detection model includes an input layer, a backbone layer, a neck layer and a prediction layer.

[0105] It is worth noting that the feature enhancement module mainly uses dilated convolution to adaptively learn the different receptive fields in each feature map according to the different scales of the detected leaves, thereby improving the accuracy of multi-scale target detection and recognition. The feature enhancement module structure can be divided into two parts: a multi-branch convolution layer and a branch pooling layer.

[0106] In the specific implementation, the multi-branch convolution layer provides receptive fields of different sizes for the input feature map through dilated convolution, and uses the average pooling layer to fuse the leaf information from the three branch receptive fields to improve the multi-scale precision prediction. The multi-branch convolution layer includes dilated convolution, batch normalization (BN) layer and ReLU activation layer. The dilated convolutions in the three parallel branches have the same kernel size, but different dilation rates. The kernel of each dilated convolution is 3×3, and the dilation rates d of different branches are 3, 5, and 7 respectively. Dilated convolution supports exponentially expanded receptive fields without losing resolution. In the convolution operation of dilated convolution, the elements of the convolution kernel are spaced, and the size of the space depends on the expansion rate. This is different from the standard convolution operation in which the elements of the convolution kernel are adjacent. The convolution kernel is changed from 3×3 to 7×7, and the receptive field of this layer is 7×7. The receptive field formula of the dilated convolution is:

[0107] r1=d×(k-1)+1

[0108] r n =d×(k-1)+r n-1

[0109] Among them, k and r represent the convolution kernel size and dilation rate respectively, and d represents the convolution step size.

[0110] The branch pooling layer is used to fuse information from different parallel branches to avoid introducing additional parameters. During the training process, the average operation is used to balance the representation of different parallel branches so that a single branch can achieve reasoning during the test process. The expression is as follows:

[0111]

[0112] Among them, y p represents the output of the branch pooling layer, B represents the number of parallel branches, and in this embodiment, B=3.

[0113] like Figure 6 As shown, Figure 6This is a schematic diagram of the feature enhancement module structure in the pesticide delivery method of this embodiment. The feature enhancement module in the figure includes a multi-branch convolution layer and a branch pooling layer. The multi-branch convolution layer includes an expanded convolution, a batch normalization (BN) layer, and a ReLU activation layer. The expanded convolutions in the three parallel branches have the same kernel size but different expansion rates. The kernel of each expanded convolution is 3×3, and the expansion rates d of different branches are 3, 5, and 7, respectively. The branch pooling layer is used to fuse information from different parallel branches.

[0114] Step S04: Improve the grape pest and disease detection model Yolov5 through iterative training of the training set until convergence and obtain the optimal network weight.

[0115] The Yolov5 detection model for grape pests and diseases was improved by training the training set, and the loss function was calculated. The network weights were calculated according to the loss function, and the optimal network weights were screened out from the network weights.

[0116] Furthermore, in order to obtain the optimal network weight, the step S04 includes: inputting the training set into the backbone layer through the input layer for feature extraction to obtain the feature map of the training set; performing a top-down downsampling operation and fusion on the feature map of the training set through the neck layer to obtain the feature map of the training set; performing prediction in the prediction layer based on the feature map of the training set, calculating the loss, and updating the model parameters; retraining until the loss function converges or the maximum number of iterations is reached to obtain the optimal network weight.

[0117] It should be noted that after the training set is preprocessed and data enhanced in the output layer, a data-enhanced training set is obtained, and the data-enhanced training set is input into the backbone layer for feature extraction to obtain a feature map of the training set.

[0118] It is understandable that preprocessing includes proportional image color adjustment of the chroma, sharpness, and brightness of all images in the training set. Since the targets of grape pests and diseases are small and inconspicuous, and the pixel area is too large, the prediction efficiency will be reduced.

[0119] Furthermore, the feature map of the training set is downsampled and fused from top to bottom through the neck layer to obtain a feature map of the training set, including: generating a sixth feature map through the adaptive attention module on the fifth feature map, fusing it with the fifth feature map and inputting it into the feature enhancement module for feature enhancement to obtain a first feature map; generating a seventh feature map through a downsampling operation on the first feature map, fusing it with the fourth feature map and inputting it into the feature enhancement module for feature enhancement to obtain a second feature map; generating an eighth feature map through a downsampling operation on the second feature map, fusing it with the third feature map and inputting it into the feature enhancement module for feature enhancement to obtain a third feature map.

[0120] It should be noted that the feature map of the training set is {C1, C2, C3, C4, C5}, including the first feature map C1, the second feature map C2, the third feature map C3, the fourth feature map C4 and the fifth feature map C5, and the feature graph of the training set includes the first feature map P5, the second feature map P4 and the third feature map P3.

[0121] It can be understood that the fifth feature map C5 is generated by the adaptive attention module AAM to generate the sixth feature map M5, which is fused with the fifth feature map C5 and input into the feature enhancement module FEM for feature enhancement to obtain the first feature map P5, the first feature map P5 is generated by the downsampling operation to generate the seventh feature map M4, which is fused with the fourth feature map C4 and input into the feature enhancement module FEM for feature enhancement to obtain the second feature map P4; the second feature map P4 is generated by the downsampling operation to generate the eighth feature map M3, which is fused with the third feature map C3 and input into the feature enhancement module FEM for feature enhancement to obtain the third feature map P3.

[0122] In its implementation, the input image generates feature maps {C1, C2, C3, C4, C5} through multiple convolutions. C5 generates feature map M6 through the adaptive attention module. M6 is summed with M5 and fused with other features from lower layers through a top-down propagation pathway. PANet shortens the information path between bottom-level and top-level features by expanding the receptive field and performing finite element analysis after each fusion. C5 serves as the input to the adaptive attention module.

[0123] like Figure 7 As shown, Figure 7This is a schematic diagram of the improved feature pyramid structure in the pesticide delivery method of this embodiment. An adaptive attention module AAM is added to the figure. The pixels of the input image are 608×608. The input image generates feature maps {C1, C2, C3, C4, C5} through multiple convolutions. C5 generates feature map M6 through the adaptive attention module. M6 is summed with M5 and propagated through a top-down pathway to fuse with other lower-level features to generate feature maps P5, P4, and P3.

[0124] Furthermore, the fifth feature map is generated into a sixth feature map through the adaptive attention module, including: obtaining a preset number of semantic features according to the fifth feature map through the adaptive pooling layer in the adaptive attention module, wherein the scale of each semantic feature in the preset number of semantic features is different; performing a convolution operation on the preset number of semantic features to convert the preset number of semantic features into the same channel dimension; performing an upsampling operation on the preset number of semantic features, and fusing the preset number of semantic features through the Concat layer to obtain a target feature map; obtaining the spatial weight of the target feature map; separating the spatial weight and the target feature map after a Hadamard product operation to obtain a preset number of new semantic features; performing a matrix addition operation on the preset number of new semantic features and the fifth feature map to obtain a sixth feature map.

[0125] It should be noted that the preset number is 3, and the scales of each semantic feature in the preset number of semantic features are β1×S, β2×S, and β3×S respectively.

[0126] In the specific implementation, C5 is used as the input of the adaptive attention module, and its size is S = H × W. First, semantic features of different scales (β1×S, β2×S, β3×S) are obtained through the adaptive pooling layer; then, each context feature is convolved with 1×1 to obtain the same channel dimension of 256; they are upsampled to the S scale using bilinear interpolation for subsequent fusion, and the spatial attention mechanism merges the channels of the three context features through the Concat layer to obtain the target feature map; the target feature map passes through the 1×1 convolution layer, ReLU activation layer, 3×3 convolution layer and sigmoid activation layer in sequence to generate corresponding spatial weights for each feature map; the generated weight map and the feature map after merging the channels are separated and added to the input feature map M5 through the Hadamard product operation, and the context features are aggregated into M6.

[0127] like Figure 8 As shown, Figure 8This is a schematic diagram of the structure of the adaptive attention module in the pesticide delivery method of this embodiment. In the figure, C5 serves as the input of the adaptive attention module, and its size is S = H × W. First, semantic features of different scales (β1×S, β2×S, β3×S) are obtained through the adaptive pooling layer. Then, each context feature is convolved 1×1 to obtain the same channel dimension of 256. They are upsampled to the S scale using bilinear interpolation for subsequent fusion. The spatial attention mechanism merges the channels of the three context features through the Concat layer; then the feature map passes through the 1×1 convolution layer, the ReLU activation layer, the 3×3 convolution layer, and the sigmoid activation layer in sequence to generate corresponding spatial weights for each feature map. The generated weight map and the feature map after merging the channels are separated and added to the input feature map M5 through the Hadamard product operation, and the context features are aggregated into M6.

[0128] Step S05: inputting the optimal network weights into the improved grape pest and disease Yolov5 detection model, optimizing the improved grape pest and disease Yolov5 detection model, and obtaining an optimized improved grape pest and disease Yolov5 detection model.

[0129] It can be understood that the Yolov5 detection model for grape diseases and pests is improved by optimizing the optimal network weights, and prediction is performed through the optimized improved Yolov5 detection model for grape diseases and pests, thereby improving the accuracy of prediction.

[0130] This embodiment obtains leaf images of grape pests and diseases and performs data amplification on the leaf images to obtain a data set; the data set is divided into a training set and a validation set according to a preset ratio; a Yolov5 detection model for grape pests and diseases is constructed, and an adaptive attention module and a feature enhancement module are added to the feature pyramid structure of the Yolov5 detection model to obtain an improved Yolov5 detection model for grape pests and diseases; the improved Yolov5 detection model for grape pests and diseases is iteratively trained on the training set until convergence to obtain optimal network weights; the optimal network weights are input into the improved Yolov5 detection model for grape pests and diseases, and the improved Yolov5 detection model for grape pests and diseases is optimized to obtain an optimized improved Yolov5 detection model for grape pests and diseases. In this manner, the Yolov5 detection model for grape pests and diseases is improved and trained by adding the adaptive attention module and the feature enhancement module to obtain optimal network weights, thereby optimizing the improved model, improving the accuracy of model detection, and improving efficiency.

[0131] Reference Figure 9 , Figure 9 This is a structural block diagram of the first embodiment of the pesticide delivery device of the present invention.

[0132] like Figure 9 As shown, the pesticide delivery device proposed in the embodiment of the present invention includes:

[0133] Detection module 10 is used to collect grape leaf images in the vineyard in real time, and detect the grape leaf images using the grape pest and disease detection model Yolov5 to obtain pest and disease data, wherein the pest and disease data includes the number of leaves in the vineyard with grape pests and diseases and the types of pests and diseases to which different leaves belong;

[0134] An execution model 20 is configured to execute a preset pesticide spraying strategy based on the pest and disease data and record reference data, wherein the reference data includes the time of pesticide spraying, the recovery of grape leaf pests and diseases in the sprayed area, and the pest and disease characteristics of each grape leaf treated with the pesticide;

[0135] An adjustment module 30 is configured to gradually adjust the preset pesticide spraying strategy according to the reference data to obtain a target pesticide spraying strategy, wherein the pesticide spraying strategy includes the concentration, type, and interval of the sprayed pesticide;

[0136] The establishment module 40 is used to establish a pesticide delivery model according to the target pesticide spraying strategy and reference data to achieve automatic pesticide delivery.

[0137] This embodiment collects grape leaf images from a vineyard in real time, detects the grape leaf images using the grape pest and disease Yolov5 detection model, and obtains pest and disease data. A preset pesticide spraying strategy is executed based on the pest and disease data and reference data is recorded. The preset pesticide spraying strategy is gradually adjusted based on the reference data to obtain a target pesticide spraying strategy. A pesticide delivery model is established based on the target pesticide spraying strategy and the reference data to achieve automatic pesticide delivery. Through the above method, pest and disease data is obtained by detecting grape leaf images using a detection model, thereby determining a target pesticide spraying strategy and establishing a pesticide delivery model to achieve automatic pesticide delivery. This solves the problem that traditional methods require a lot of time and effort and have low accuracy, achieving automation, improving efficiency, and achieving high delivery accuracy.

[0138] In one embodiment, the establishment module 40 is also used to store the pest and disease data and the reference data as several experience samples, and calculate the accuracy and recall rate of the data in each experience sample; and train the pesticide delivery model according to the accuracy and recall rate of the data in each experience sample.

[0139] In one embodiment, the adjustment module 30 is further configured to increase or decrease the concentration of the sprayed pesticide according to the recovery of grape leaf diseases and insect pests in the sprayed area until an optimal spraying concentration is determined; increase or decrease the time interval for spraying the pesticide according to the recovery of grape leaf diseases and insect pests in the sprayed area until an optimal spraying time interval is determined; select several types of spraying pesticides for alternating use according to the recovery of grape leaf diseases and insect pests in the sprayed area; and use the optimal spraying concentration of the pesticide, the optimal spraying time interval of the pesticide, and the several types of spraying pesticides as the target pesticide spraying strategy.

[0140] In one embodiment, the detection module 10 is further configured to obtain images of grape pests and diseases and perform data amplification on the images of grape pests and diseases to obtain a data set; divide the data set into a training set and a validation set according to a preset ratio; construct a Yolov5 detection model for grape pests and diseases, and add an adaptive attention module and a feature enhancement module to the feature pyramid structure of the Yolov5 detection model for grape pests and diseases to obtain an improved Yolov5 detection model for grape pests and diseases, wherein the Yolov5 detection model for grape pests and diseases includes an input layer, a backbone layer, a neck layer, and a prediction layer; iteratively train the improved Yolov5 detection model for grape pests and diseases through the training set until convergence to obtain an optimal network weight; input the optimal network weight into the improved Yolov5 detection model for grape pests and diseases, and optimize the improved Yolov5 detection model for grape pests and diseases to obtain an optimized improved Yolov5 detection model for grape pests and diseases.

[0141] In one embodiment, the detection module 10 is further used to input the training set into the backbone layer through the input layer for feature extraction to obtain a feature map of the training set; perform top-down downsampling and fusion on the feature map of the training set through the neck layer to obtain a feature map of the training set; perform prediction in the prediction layer based on the feature map of the training set, calculate the loss, and update the model parameters; retrain until the loss function converges or the maximum number of iterations is reached to obtain the optimal network weights.

[0142] In one embodiment, the feature map of the training set includes a first feature map, a second feature map, a third feature map, a fourth feature map and a fifth feature map; the feature graph of the training set includes a first feature map, a second feature map and a third feature map, and the detection module 10 is further used to generate a sixth feature map from the fifth feature map through the adaptive attention module, and fuse it with the fifth feature map and input it into the feature enhancement module for feature enhancement to obtain a first feature map; generate a seventh feature map from the first feature map through a downsampling operation, and fuse it with the fourth feature map and input it into the feature enhancement module for feature enhancement to obtain a second feature map; generate an eighth feature map from the second feature map through a downsampling operation, and fuse it with the third feature map and input it into the feature enhancement module for feature enhancement to obtain a third feature map.

[0143] In one embodiment, the detection module 10 is further configured to obtain a preset number of semantic features through an adaptive pooling layer in the adaptive attention module according to the fifth feature map, wherein each of the preset number of semantic features has a different scale;

[0144] A convolution operation is performed on the preset number of semantic features to convert the preset number of semantic features into the same channel dimension; an upsampling operation is performed on the preset number of semantic features, and the preset number of semantic features are fused through a Concat layer to obtain a target feature map; the spatial weight of the target feature map is obtained; the spatial weight and the target feature map are separated after a Hadamard product operation to obtain a preset number of new semantic features; a matrix addition operation is performed on the preset number of new semantic features and the fifth feature map to obtain a sixth feature map.

[0145] In addition, to achieve the above-mentioned purpose, the present invention also proposes a pesticide delivery device, which includes: a memory, a processor, and a pesticide delivery program stored on the memory and executable on the processor, wherein the pesticide delivery program is configured to implement the steps of the pesticide delivery method described above.

[0146] Since the pesticide delivery device adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be described one by one here.

[0147] In addition, an embodiment of the present invention further provides a storage medium on which a pesticide delivery program is stored. When the pesticide delivery program is executed by a processor, the steps of the pesticide delivery method described above are implemented.

[0148] Since the storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought by the technical solutions of the above embodiments, which will not be described one by one here.

[0149] It should be understood that the above is only an example and does not constitute any limitation to the technical solution of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any limitation on this.

[0150] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of the embodiment scheme, and no limitation is made here.

[0151] In addition, for technical details not fully described in this embodiment, reference can be made to the pesticide delivery method provided in any embodiment of the present invention, and will not be repeated here.

[0152] In addition, it should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0153] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0154] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, or of course by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0155] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for applying pesticides, characterized in that: The method comprises: Real-time collection of grape leaf images in the vineyard, detection of the grape leaf images using the grape pest and disease Yolov5 detection model, and obtaining pest and disease data, wherein the pest and disease data includes the number of leaves in the vineyard with grape pests and diseases and the types of pests and diseases to which different leaves belong; Executing a preset pesticide spraying strategy based on the pest and disease data and recording reference data, wherein the reference data includes the time of pesticide spraying, the recovery of grape leaf pests and diseases in the sprayed area, and the pest and disease characteristics of each grape leaf treated with the pesticide; Stepwise adjusting the preset pesticide spraying strategy according to the reference data to obtain a target pesticide spraying strategy, wherein the pesticide spraying strategy includes the concentration, type and interval of the spraying pesticide; Establishing a pesticide delivery model based on the target pesticide spraying strategy and reference data to achieve automatic pesticide delivery; Before collecting grape leaf images in the vineyard in real time and detecting the grape leaf images using the grape pest and disease Yolov5 detection model to obtain pest and disease data, the method further includes: Obtaining leaf images of grape diseases and insect pests, and performing data amplification on the leaf images of grape diseases and insect pests to obtain a data set; Divide the data set into a training set and a validation set according to a preset ratio; Constructing a Yolov5 detection model for grape pests and diseases, and adding an adaptive attention module and a feature enhancement module to the feature pyramid structure of the Yolov5 detection model to obtain an improved Yolov5 detection model for grape pests and diseases, wherein the Yolov5 detection model for grape pests and diseases includes an input layer, a backbone layer, a neck layer, and a prediction layer; Improve the Yolov5 grape pest and disease detection model through iterative training of the training set until convergence and obtain the optimal network weights; Inputting the optimal network weights into the improved Yolov5 detection model for grape diseases and insect pests, optimizing the improved Yolov5 detection model for grape diseases and insect pests, and obtaining an optimized improved Yolov5 detection model for grape diseases and insect pests; The method of improving the grape pest and disease Yolov5 detection model by iterative training of the training set until convergence to obtain the optimal network weight includes: Inputting the training set into the backbone layer through the input layer for feature extraction to obtain a feature map of the training set; Performing a top-down downsampling operation and a fusion operation on the feature map of the training set through the neck layer to obtain a feature map of the training set; Perform prediction in the prediction layer based on the feature map of the training set, calculate the loss, and update the model parameters; Retrain until the loss function converges or the maximum number of iterations is reached to obtain the optimal network weights.

2. The method according to claim 1, wherein After establishing a pesticide delivery model based on the target pesticide spraying strategy and reference data to achieve automatic pesticide delivery, the method further includes: Storing the pest and disease data and the reference data as a plurality of experience samples, and calculating the accuracy and recall of the data in each experience sample; The pesticide delivery model is trained according to the accuracy and recall of the data in each experience sample.

3. The method according to claim 1, wherein The step of gradually adjusting the preset pesticide spraying strategy according to the reference data to obtain a target pesticide spraying strategy includes: Increase or decrease the concentration of sprayed pesticides according to the recovery of grape leaf diseases and pests in the sprayed area until the optimal concentration of sprayed pesticides is determined; Increase or decrease the interval between spraying pesticides according to the recovery of grape leaf pests and diseases in the sprayed area until the optimal spraying interval is determined; Select several types of spraying pesticides for alternating use based on the recovery of grape leaf diseases and insect pests in the sprayed area; The optimal pesticide spraying concentration, the optimal pesticide spraying time interval, and the types of several pesticides to be sprayed are used as target pesticide spraying strategies.

4. The method according to claim 1, wherein The feature maps of the training set include a first feature map, a second feature map, a third feature map, a fourth feature map, and a fifth feature map; The feature graphs of the training set include a first feature graph, a second feature graph, and a third feature graph; The step of performing a top-down downsampling operation and a fusion operation on the feature map of the training set through the neck layer to obtain a feature map of the training set includes: The fifth feature map is generated by the adaptive attention module into a sixth feature map, and the sixth feature map is fused with the fifth feature map and input into the feature enhancement module for feature enhancement to obtain a first feature map; Downsampling the first feature map to generate a seventh feature map, fusing the seventh feature map with the fourth feature map and inputting the seventh feature map into the feature enhancement module for feature enhancement to obtain a second feature map; The second feature map is down-sampled to generate an eighth feature map, which is fused with the third feature map and then input into the feature enhancement module for feature enhancement to obtain a third feature map.

5. The method according to claim 4, wherein The step of generating a sixth feature map by using the adaptive attention module on the fifth feature map includes: Obtaining a preset number of semantic features through an adaptive pooling layer in the adaptive attention module according to the fifth feature map, wherein each of the preset number of semantic features has a different scale; Performing a convolution operation on the preset number of semantic features to convert the preset number of semantic features into the same channel dimension; Performing an upsampling operation on the preset number of semantic features, and fusing the preset number of semantic features through a Concat layer to obtain a target feature map; Obtaining a spatial weight of the target feature map; Separating the spatial weight and the target feature map after performing a Hadamard product operation to obtain a preset number of new semantic features; A matrix addition operation is performed on the preset number of new semantic features and the fifth feature map to obtain a sixth feature map.

6. A pesticide delivery device, characterized in that: The pesticide delivery device comprises: A detection module is used to collect grape leaf images in the vineyard in real time, and detect the grape leaf images using the grape pest and disease Yolov5 detection model to obtain pest and disease data, wherein the pest and disease data includes the number of leaves in the vineyard with grape pests and diseases and the types of pests and diseases to which different leaves belong; an execution model for executing a preset pesticide spraying strategy based on the pest and disease data and recording reference data, wherein the reference data includes the time of pesticide spraying, the recovery of grape leaf pests and diseases in the sprayed area, and the pest and disease characteristics of each grape leaf treated with the pesticide; an adjustment module, configured to gradually adjust the preset pesticide spraying strategy according to the reference data to obtain a target pesticide spraying strategy, wherein the pesticide spraying strategy includes the concentration, type, and interval of the spraying pesticide; Establishing a module for establishing a pesticide delivery model based on the target pesticide spraying strategy and reference data to achieve automatic pesticide delivery; The detection module is further used to obtain images of grape leaves with diseases and insect pests, and perform data amplification on the images of grape leaves with diseases and insect pests to obtain a data set; Divide the data set into a training set and a validation set according to a preset ratio; Constructing a Yolov5 detection model for grape pests and diseases, and adding an adaptive attention module and a feature enhancement module to the feature pyramid structure of the Yolov5 detection model to obtain an improved Yolov5 detection model for grape pests and diseases, wherein the Yolov5 detection model for grape pests and diseases includes an input layer, a backbone layer, a neck layer, and a prediction layer; Improve the Yolov5 grape pest and disease detection model through iterative training of the training set until convergence and obtain the optimal network weights; Inputting the optimal network weights into the improved Yolov5 detection model for grape diseases and insect pests, optimizing the improved Yolov5 detection model for grape diseases and insect pests, and obtaining an optimized improved Yolov5 detection model for grape diseases and insect pests; The detection module is further configured to input the training set into the backbone layer through the input layer to perform feature extraction, thereby obtaining a feature map of the training set; Performing a top-down downsampling operation and a fusion operation on the feature map of the training set through the neck layer to obtain a feature map of the training set; Perform prediction in the prediction layer based on the feature map of the training set, calculate the loss, and update the model parameters; Retrain until the loss function converges or the maximum number of iterations is reached to obtain the optimal network weights.

7. A pesticide delivery device, characterized in that: The pesticide delivery device includes: a memory, a processor, and a pesticide delivery program stored in the memory and executable on the processor, wherein the pesticide delivery program is configured to implement the pesticide delivery method according to any one of claims 1 to 5.

8. A storage medium, characterized in that: The storage medium stores a pesticide delivery program, which, when executed by the processor, implements the pesticide delivery method according to any one of claims 1 to 5.

Images