Pepper insect pest image recognition system and recognition method

Abstract

The invention discloses a pepper insect pest image recognition system and recognition method. The system comprises a camera module, a storage module, an analysis module, an inertial navigation module and a pesticide application module, acquiring image information and dimension information of insect pests through a camera module; image information and dimension information are stored through a storage module; calculating and obtaining an insect pest distribution map through an analysis module, dividing the distribution map into different density areas, marking the different density areas, and obtaining density identifiers; the inertial navigation module is used for collecting vehicle data and sending the collected data to the analysis module so as to adjust the pose of the vehicle in real time; the pesticide application module is used for executing different pesticide application instructions based on different density areas through combination of the inertial navigation module and the analysis module. According to the invention, image recognition is carried out through the preset model, the type and position information of pepper pests is accurately recognized, and the recognition accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the field of agricultural technology, and relates to a pepper pest image recognition system and recognition method, and specifically to a system and method for accurately identifying common pepper pests and accurately applying pesticides. Background Art

[0002] With the development of science and technology, agricultural robots and intelligent pesticide application systems based on image recognition and intelligent control technology are gradually being applied in agricultural production. Pest image recognition is crucial in the agricultural field. It can help farmers find and identify pests in a timely manner, thereby reducing damage to crops. Through automated monitoring, the image recognition system can improve the accuracy and efficiency of recognition and reduce labor costs. In addition, it can also support precise pesticide application, optimize pesticide use, reduce environmental impact, and promote sustainable agricultural development, thereby helping to increase crop yields and quality.

[0003] At present, pest identification mainly adopts the following technologies: Image processing and feature extraction: Pests are identified through traditional image processing methods, such as edge detection and color segmentation. However, these methods are sensitive to changes in lighting and background, and have low recognition accuracy.

[0004] Machine learning: Use traditional machine learning algorithms for classification, which usually requires manual feature extraction, is inefficient, and performs poorly when processing complex data.

[0005] Deep learning: Using convolutional neural networks (CNN) for pest identification has become the mainstream. This method can automatically extract features and has a high recognition accuracy. However, it has high requirements on the quality and quantity of training data, and the training process is computationally intensive, resulting in high time and cost.

[0006] At present, there are still few pest identification and precision pesticide application systems for pepper cultivation on the market, and there are problems such as low identification accuracy and inaccurate pesticide application. Therefore, there is an urgent need for a system and method that can efficiently and accurately identify common pepper pests and apply pesticides accurately to improve the utilization rate of pesticides, reduce environmental pollution, and ensure the safety of agricultural products. Summary of the invention

[0007] The purpose of the present invention is to address the deficiencies in the above-mentioned prior art and propose a method for pepper pest image recognition and targeted pesticide application to improve the efficiency and yield of pepper planting, reduce the amount of pesticide used, reduce the risk of environmental pollution, and ensure the safety of agricultural products.

[0008] The pepper pest image recognition system provided in this application adopts the following technical solutions: A pepper pest image recognition system, comprising a frame; characterized in that the pest recognition system also includes: The camera module is composed of an industrial camera, a depth camera and a data transmission line. The image information of the pests is obtained by the industrial camera, and the dimensional information of the pests is obtained by the depth camera. A storage module, composed of a readable storage medium and a data sending device, for storing the image information and dimensional information, and sending the data; An analysis module, composed of a computer and a data receiving device, the data receiving device is connected to the data sending device, the computer obtains a pest distribution map based on the dimensional information, divides the distribution map into different density areas, marks the different density areas, and obtains density identifiers; The inertial navigation module is composed of a data transmission device, a speed sensor, an inertial sensor and a positioning device. The inertial navigation module collects vehicle data and sends the collected data to the analysis module to adjust the vehicle posture in real time. The medicine application module is composed of a medicine box, an air inlet pipe, a medicine outlet pipe and a solenoid valve nozzle. The medicine application module executes different medicine application instructions based on different density areas through the combination of the inertial navigation module and the analysis module.

[0009] Furthermore, the dimensional information is composed of: the first dimensional information represents the horizontal height information of the pepper pests, and the second dimensional information represents the straight-line distance information between the pepper pests and the sampling point.

[0010] Furthermore, the density mark includes a first density mark and a second density mark, the first density mark indicates that the distribution density of pepper pests is lower than a predetermined value of one, and the second density mark indicates that the distribution density of pepper pests is lower than a predetermined value of two.

[0011] Furthermore, the drug administration instruction includes a first concentration instruction and a second concentration instruction; the first density identifier corresponds to the first concentration instruction, and the second density identifier corresponds to the second concentration instruction.

[0012] By adopting the above technical solution, the above-mentioned first dimensional information and second dimensional information, the first density identifier and the second density identifier, the first concentration instruction and the second concentration instruction are in a progressive relationship. The location of the pests is recorded by the two-dimensional information. When the pest density in a certain block reaches a certain number, it is marked with the first density identifier. At the same time, the first density identifier corresponds to the first concentration of pesticide application instruction. Similarly, when the number of pests in a certain block reaches the threshold of the second concentration identifier, the triggered pesticide application instruction is the second concentration of pesticide application instruction. By spraying different concentrations of drugs in areas with different pest concentrations, the utilization rate of pesticides can be improved and the pollution of pesticides to the environment can be reduced.

[0013] The present application provides a pepper pest image recognition method using the following technical solutions: S1. Image acquisition: The image of the pest to be identified is acquired by an industrial camera and transmitted to the analysis module computer through a data transmission device for further analysis and processing; S2. Image preprocessing: The preset model preprocesses the input image and adjusts the input image to 640×640 pixels to ensure input consistency, thereby speeding up calculations and reducing complexity; S3, Feature Extraction: After preprocessing, the pest image to be identified is extracted through a convolutional neural network to extract image features. Through multiple convolutional layers and activation functions, the edge, texture, shape and other information in the image are captured; S4. Generate anchor boxes: After feature recognition, the preset model generates anchor boxes. Anchor boxes are predefined rectangular boxes on the feature map. The model uses these boxes to detect objects of different sizes and shapes. The size of the anchor boxes is determined by the clustering algorithm to adapt to the diversity of objects in the training data set. S5, Bounding box regression: For each anchor box, the preset model uses a regression method to adjust the position and size of the box to better fit the actual target, and calculates the difference between the anchor box and the actual target box to achieve model learning and how to accurately locate the target; S6. Category prediction: After the anchor box generation is completed, the preset model will perform category prediction. Each anchor box not only predicts the position, but also predicts the target category. Through the fully connected layer, the model outputs the category probability distribution of each box, and uses the softmax function to convert the output into probability, so that the sum of the probabilities of all categories is 1; S7, non-maximum suppression: After the prediction is completed, the preset model will perform non-maximum suppression. Non-maximum suppression is a technology used to eliminate overlapping detections. By setting a confidence threshold and an IOU (Intersection over Union) threshold, the box with the highest confidence is retained and the boxes that overlap with it by more than the threshold are removed, thereby ensuring the uniqueness of the final result. S8. Output results: Finally, the preset model outputs the category, bounding box coordinates and confidence score of each target. According to the type information output by the model, the high-performance computer divides the identified pest images, generates classified images, and adds corresponding type information to each classified image, which provides effective support for subsequent ground pest prediction and prevention.

[0014] In summary, the present invention includes at least one of the following beneficial technical effects: (1) The present invention performs image recognition through a preset model, accurately identifies the type and location information of pepper pests, and improves the accuracy of recognition.

[0015] (2) Based on the pest distribution map and density mark, the present invention makes pesticide application more accurate, avoids over- or under-application of pesticides, and improves the utilization rate of pesticides.

[0016] (3) The automated identification and application system of the present invention can reduce the labor intensity of operators and further improve work efficiency.

[0017] (4) The present invention can reduce the amount of pesticides used, lower the risk of environmental pollution, and further ensure the safety of agricultural products. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the system composition of the present invention.

[0019] Figure 2 This is a diagram of the training effect of the preset model of the present invention.

[0020] Figure 3 This is a diagram of the training effect of the preset model of the present invention.

[0021] Figure 4 This is a preset model training parameter map for the present invention.

[0022] Figure 5 The figure is a flow chart of the pest image recognition method of the present invention.

[0023] Figure 6 The present invention is a flow chart of the concentration identification method. DETAILED DESCRIPTION

[0024] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art fall within the scope defined by the claims attached to this application.

[0025] like Figure 1As shown, a pepper pest image recognition system includes a frame, a camera module, a storage module, an analysis module, an inertial navigation module and a pesticide application module arranged on the frame; the camera module is composed of an industrial camera, a depth camera and a data transmission line, and the image information of the pests is obtained through the industrial camera, and the dimensional information of the pests is obtained through the depth camera; the storage module is composed of a readable storage medium and a data sending device, and is used to store image information and dimensional information, and send the data; the analysis module is composed of a computer and a data receiving device, and the data receiving device is connected to the data sending device. The computer obtains a pest distribution map based on the dimensional information, divides the distribution map into different density areas, marks the different density areas, and obtains density identification; the inertial navigation module is composed of a data sending device, a speed sensor, an inertial sensor and a positioning device, and the vehicle data is collected through the inertial navigation module, and the collected data is sent to the analysis module to adjust the vehicle posture in real time; the pesticide application module is composed of a medicine box, an air intake pipe, a medicine outlet pipe and a solenoid valve nozzle. The pesticide application module executes different pesticide application instructions based on different density areas through the combination of the inertial navigation module and the analysis module.

[0026] Specifically, Figure 1 As shown, the dimensional information is composed of the first dimensional information and the second dimensional information. The first dimensional information represents the horizontal height information of pepper pests, and the second dimensional information represents the straight-line distance information between the pepper pests and the sampling point. The density mark includes the first density mark and the second density mark. The first density mark represents that the distribution density of pepper pests is lower than the predetermined value one, and the second density mark represents that the distribution density of pepper pests is lower than the predetermined value two. The air intake pipeline is connected and arranged at the top of the medicine box, and the medicine outlet pipeline is connected and arranged at the bottom of the medicine box. The other end of the medicine outlet pipeline is connected to the solenoid valve nozzle. The pesticide application instruction includes the first concentration instruction and the second concentration instruction; the first density mark corresponds to the first concentration instruction, and the second density mark corresponds to the second concentration instruction.

[0027] like Figure 5-6 As shown, a pepper pest image recognition method is as follows: S1. Image acquisition: The image of the pest to be identified is acquired by an industrial camera and transmitted to the analysis module computer through a data transmission device for further analysis and processing; S2. Image preprocessing: The preset model preprocesses the input image and adjusts the input image to 640×640 pixels to ensure input consistency, thereby speeding up calculations and reducing complexity; S3, Feature Extraction: After preprocessing, the pest image to be identified is extracted through a convolutional neural network to extract image features. Through multiple convolutional layers and activation functions, the edge, texture, shape and other information in the image are captured; S4. Generate anchor boxes: After feature recognition, the preset model generates anchor boxes. Anchor boxes are predefined rectangular boxes on the feature map. The model uses these boxes to detect objects of different sizes and shapes. The size of the anchor boxes is determined by the clustering algorithm to adapt to the diversity of objects in the training data set. S5, Bounding box regression: For each anchor box, the preset model uses a regression method to adjust the position and size of the box to better fit the actual target, and calculates the difference between the anchor box and the actual target box to achieve model learning and how to accurately locate the target; S6. Category prediction: After the anchor box generation is completed, the preset model will perform category prediction. Each anchor box not only predicts the position, but also predicts the target category. Through the fully connected layer, the model outputs the category probability distribution of each box, and uses the softmax function to convert the output into probability, so that the sum of the probabilities of all categories is 1; S7, non-maximum suppression: After the prediction is completed, the preset model will perform non-maximum suppression. Non-maximum suppression is a technology used to eliminate overlapping detections. By setting a confidence threshold and IOU threshold, the box with the highest confidence is retained and the box that overlaps with it by more than the threshold is removed, thereby ensuring the uniqueness of the final result. S8. Output results: Finally, the preset model outputs the category, bounding box coordinates and confidence score of each target. According to the species information output by the model, the high-performance computer divides the identified pest images, generates classified images, and adds corresponding species information to each classified image, which provides effective support for subsequent ground pest prediction and prevention.

[0028] like Figure 2-3 As shown, after the image is transmitted to the preset model, it will be processed by the model and the category of each target will be output. According to the category information output by the model, the high-performance computer will perform further processing.

[0029] like Figure 4 As shown in the figure, the training effect of the preset model is visualized. The parameter shown in figure a is the prediction accuracy (Precision), that is, the ratio of the number of samples correctly predicted as positive categories to the number of all samples predicted as positive categories; the parameter shown in figure b is the recall rate (Recall), that is, the ratio of the number of samples correctly predicted as positive categories to the number of all true positive samples; the parameter shown in figure c is the average precision (mAP50), that is, the area calculated using the Precision-Recall curve, which can comprehensively reflect the accuracy of model prediction; the parameter shown in figure d is the average precision (mAP50-95): it represents the average mAP under different IoU thresholds. Compared with mAP50, its numerical requirements are more stringent.

[0030] Furthermore, the pepper pest location information collected by the depth camera is input into a preset model for analysis, and distribution identification is performed based on the pepper pest location information to obtain first dimensional information and second dimensional information.

[0031] Combined with the pest type information, the first dimension information and the second dimension information of each pepper pest are input into the distribution map through the density analysis module to obtain the density distribution information of the pepper pest. Based on the density distribution information, the distribution map is divided into different density areas. The different density areas divided based on the distribution map are marked to obtain density identifiers, including at least a first density identifier and a second density identifier. The first density identifier indicates that the distribution density of pepper pests is lower than a predetermined value of one, and the second density identifier indicates that the distribution density of pepper pests is lower than a predetermined value of two. The high-performance computer sends different pesticide application instructions to the pesticide application device.

[0032] Furthermore, the targeted drug application device executes different drug application instructions for different marked areas during movement according to different drug application instructions issued by the computer and in combination with the position of the vehicle body, and triggers different targeted drug application instructions based on different density areas.

[0033] Furthermore, the high-performance computer processes the three-dimensional characteristics of the environment and the posture and displacement information of the equipment according to the vehicle's real-time position information, vehicle speed information and vehicle inertia information, updates and adjusts the vehicle's posture in real time, adjusts the solenoid valve switching time, and achieves precise targeted application of pesticides.

[0034] The pepper pest targeted pesticide application device of the present invention can achieve accurate positioning and application of pepper pests. The combination of the density analysis module and the targeted pesticide application module ensures the pertinence and effectiveness of the pesticide application, reduces the waste of pesticides and environmental pollution, and improves the effect of pest control. The fusion positioning of the inertial navigation module and the depth camera module ensures the stability and accuracy of the equipment in a dynamic environment, further improving the accuracy and efficiency of targeted pesticide application.

Claims

1. A pepper pest image recognition system, comprising a frame; characterized in that: The pest identification system further includes: The camera module is composed of an industrial camera, a depth camera and a data transmission line. The image information of the pests is obtained by the industrial camera, and the dimensional information of the pests is obtained by the depth camera. A storage module, composed of a readable storage medium and a data sending device, for storing the image information and dimensional information, and sending the data; An analysis module, composed of a computer and a data receiving device, the data receiving device is connected to the data sending device, the computer obtains a pest distribution map based on the dimensional information, divides the distribution map into different density areas, marks the different density areas, and obtains density identifiers; The inertial navigation module is composed of a data transmission device, a speed sensor, an inertial sensor and a positioning device. The inertial navigation module collects vehicle data and sends the collected data to the analysis module to adjust the vehicle posture in real time. The medicine application module is composed of a medicine box, an air inlet pipe, a medicine outlet pipe and a solenoid valve nozzle. The medicine application module executes different medicine application instructions based on different density areas through the combination of the inertial navigation module and the analysis module.

2. A pepper pest image recognition system according to claim 1, characterized in that: The dimensional information is composed of first dimensional information and second dimensional information, the first dimensional information represents the horizontal height information of pepper pests, and the second dimensional information represents the straight-line distance information between the pepper pests and the sampling point.

3. A pepper pest image recognition system according to claim 1, characterized in that: The density mark includes a first density mark and a second density mark, the first density mark indicates that the distribution density of pepper pests is lower than a predetermined value of one, and the second density mark indicates that the distribution density of pepper pests is lower than a predetermined value of two.

4. The pepper pest image recognition system according to claim 1, characterized in that: The drug administration instruction includes a first concentration instruction and a second concentration instruction; the first density identifier corresponds to the first concentration instruction, and the second density identifier corresponds to the second concentration instruction.

5. A pepper pest image recognition method, characterized in that: Using the pepper pest image recognition system according to any one of claims 1 to 4, the method is as follows: S1. Image acquisition: The image of the pest to be identified is acquired by an industrial camera and transmitted to the analysis module computer through a data transmission device for further analysis and processing; S2. Image preprocessing: The preset model preprocesses the input image and adjusts the input image to 640×640 pixels to ensure input consistency, thereby speeding up calculations and reducing complexity; S3, Feature Extraction: After preprocessing, the pest image to be identified is extracted through a convolutional neural network to extract image features. Through multiple convolutional layers and activation functions, the edge, texture, shape and other information in the image are captured; S4, Generate anchor boxes: After completing feature recognition, the preset model generates anchor boxes. The model uses these anchor boxes to detect objects of different sizes and shapes. The size of the anchor boxes is determined by the clustering algorithm to adapt to the diversity of objects in the training data set. S5, Bounding box regression: For each anchor box, the preset model uses a regression method to adjust the position and size of the box to better fit the actual target, and calculates the difference between the anchor box and the actual target box to achieve model learning and how to accurately locate the target; S6, Category prediction: After the anchor box generation is completed, the preset model will perform category prediction. The anchor box predicts the position and target category. Through the fully connected layer, the model outputs the category probability distribution of each box, and uses the softmax function to convert the output into probability so that the sum of the probabilities of all categories is 1; S7, non-maximum suppression: After the prediction is completed, the preset model will perform non-maximum suppression. Non-maximum suppression is a technology used to eliminate overlapping detections. By setting a confidence threshold and IOU threshold, the box with the highest confidence is retained and the box that overlaps with it by more than the threshold is removed, thereby ensuring the uniqueness of the final result. S8. Output results: The preset model outputs the category, bounding box coordinates and confidence score of each target. According to the type information output by the model, the computer divides the identified pest images, generates classified images, and adds corresponding type information to each classified image to provide support for subsequent pest prediction and prevention.

Images