Dynamic detection system for insect pests of tea trees

Abstract

The invention relates to the technical field of insect pest detection, in particular to a tea tree insect pest dynamic detection system which comprises observation equipment, a regulation and control system and a tea tree insect pest detection improved model. The observation equipment is used for data processing coordination, capturing tea garden insect pest situation images and simultaneously realizing high-speed and stable data transmission; the regulation and control system provides an interactive operation interface of the remote observation equipment and is used for server connection, equipment control and picture data uploading; according to the YOLOv8-Faster Tea model provided by the system, on the basis of a YOLOv8 original model, a P5 layer is removed, a P2 small target detection layer is added, and meanwhile, an original C2f module is replaced with a partial convolution module. According to the invention, the ultra-high-definition microscopic camera is adopted, so that the observation capability is excellent; the method has the advantages of simple operation process, fast user operation, support of remote deployment and debugging, time and cost saving, high efficiency and energy saving of the original algorithm, and suitableness for various deployment environments; the pheromone board is adopted as a main observation object, and the mode is economical, practical and easy and convenient to maintain.

Description

Technical Field

[0001] The invention relates to the technical field of insect pest detection, in particular to a tea tree insect pest dynamic detection system. Background Art

[0002] The integrated pest control technology for tea trees developed by the Tea Research Institute of the Chinese Academy of Agricultural Sciences and other institutions has conducted in-depth research on three insects and one disease in tea gardens (blackthorn whitefly, tea looper, tea orange gall mite and tea white star disease), and established an integrated pest control system and decision-making system for tea trees in different ecological zones. The technical solution includes the use of biological agents for prevention and control, optimizing the use of pesticides, and establishing resistance monitoring methods. Although the above technology has achieved certain results, it may have the following disadvantages (disadvantages under traditional manual detection):

[0003] 1. The effectiveness of biological agents may be affected by environmental factors. For example, weather conditions may reduce their effectiveness.

[0004] 2. Although optimized use of pesticides reduces the number of spraying times, long-term reliance on chemical pesticides may lead to pest resistance.

[0005] 3. In addition to pest control, it is also necessary to observe the pest types and danger levels, and summarize the pest cycle and disease incidence. However, traditional manual observation is time-consuming, labor-intensive, ineffective, and discontinuous.

[0006] 4. Pest observation requires professional identification of pests and diseases, and requires a high level of experience and skill from the observer, but local areas lack such talent.

[0007] However, in general, the current tea tree pest and disease detection system based on the Internet of Things utilizes information technologies such as the Internet of Things, computer image recognition, and GIS, which shows that the existing technology has begun to develop in the direction of intelligence and automation. However, the disadvantages that this system may face include:

[0008] 1. The existing equipment mainly consists of observation equipment such as insect monitoring lights, which are expensive and difficult to install and popularize.

[0009] 2. Currently, the equipment on the market is only for rough observation, and generally only counts the number of pests in a certain period of time, but does not have the conditions to observe the pests and their disease cycles. On the one hand, the cameras of the observation equipment are cheap and the pixels are not enough. On the other hand, the algorithm recognition ability is limited and cannot improve high-precision detection.

[0010] 3. Generally, the pests collected by insect monitoring lights and other equipment are destructive to the insects themselves. For example, the wings, limbs, torso and other parts may be smashed during the trapping process.

[0011] In summary, a tea tree pest dynamic detection system needs to be proposed to solve the above problems. Summary of the invention

[0012] The object of the present invention is to provide a tea tree pest dynamic detection system to solve the problems raised in the above background technology.

[0013] To achieve the above object, the present invention provides the following technical solutions:

[0014] Tea tree pest dynamic detection system, including observation equipment, control system and tea tree pest detection improvement model;

[0015] The observation equipment is used for data processing and coordination, capturing images of tea garden insect conditions and achieving high-speed and stable data transmission;

[0016] The control system provides an interactive operation interface for remote observation equipment, which is used for server connection, equipment control and image data uploading;

[0017] The improved tea tree pest detection model is based on the Yolov8 network, using the Faster Block module to replace the C2f module, while removing the P5 target detection layer of the PAN-FPN structure in the original neck network, adding a P2 small target detection layer on the upper part of the structure to identify and locate the pest position, and the improved model is defined as YOLOv8-FasterTea.

[0018] Preferably, the observation device further comprises:

[0019] A microcomputer, wherein the microcomputer is a Raspberry Pi 4B microcomputer, which is used for data processing and coordination;

[0020] A color USB high-definition industrial camera with a resolution of 12 megapixels, used to capture images of insects in tea gardens;

[0021] 5G outdoor WIFI module for data transmission.

[0022] Preferably, the observation equipment also includes an energy system and a power supply voltage stabilization conversion module. The energy system consists of a 120W solar panel and a 240AH battery, which is used to provide energy support for the observation equipment in an outdoor environment. The power supply voltage stabilization conversion module is used for stable power supply under different voltage conditions.

[0023] Preferably, the control system adopts Vue technology framework to write the control interface, and the operation logic is implemented in Python language.

[0024] Preferably, the control system includes a server, a device control unit and a picture data uploading module;

[0025] The server uses FRP technology to realize the control of the remote control terminal (i.e., Raspberry Pi) through the pre-configured IP address through the web interface;

[0026] The device control unit uses the OpenCV library to control the camera, set the timed photo function, and reserve the light on and camera movement functions to transmit the captured insect pest images to the remote image bed server, and the image information is forwarded by the server to other servers for storage;

[0027] After obtaining the token of the image storage server, the image data is transmitted according to the specified API format.

[0028] Preferably, the improved tea tree pest detection model is used for image feature extraction, comprising the following steps:

[0029] S1. Create a partial convolution operator for spatial feature extraction. Apply conventional convolution to a portion of the input channels for spatial feature extraction, and keep the remaining channels unchanged, as shown in Equation (1) and Equation (2):

[0030] FLOPs pconv =h·w·k 2 ·c p 2 (1);

[0031] MAC pconv =h·w·2c p +k 2 ·c p 2 ≈h·w·2c p (2);

[0032] In formula (1) and formula (2), h is the height of the input feature map, w is the width of the input feature map, k is the size of the convolution kernel, and c is p is the number of input channels;

[0033] In formula (1) and formula (2), FLOPs is the amount of computation; MAC is the memory access cost, and pconv is partial convolution;

[0034] FasterNet Block sets two PWConv layers after the PConv layer and only places normalization and activation layers after the intermediate layers;

[0035] S2. Remove the P5 target detection layer of the PAN-FPN structure in the original neck network, and add a P2 small target detection layer on the upper part of the structure. The P2 small target detection layer is used to perform feature fusion on the feature maps of the first few layers in the backbone network.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention adopts an ultra-high-definition microscope camera with excellent observation capabilities; the operation process is simple, users can get started quickly, support remote deployment and debugging, save time and cost, the original algorithm is efficient and energy-saving, and is suitable for a variety of deployment environments; the sticky insect board trapping technology is economical and practical, and maintenance is simple. At the same time, the equipment of the present invention is low-cost, self-powered, networked, easy to install and operate, and plug-and-play, providing a comprehensive pest monitoring and early warning solution for tea gardens, significantly improving tea yield and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The structure diagram of the tea tree pest dynamic detection system of the present invention is shown;

[0038] Figure 2 The topological diagram of the tea tree pest dynamic detection system of the present invention is shown;

[0039] Figure 3 The control system architecture diagram of the present invention is shown;

[0040] Figure 4 The YOLOv8n network structure diagram of the present invention is shown;

[0041] Figure 5 The YOLOv8-FasterTea network structure diagram of the present invention is shown;

[0042] Figure 6 The operation flow chart of the present invention is shown. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] See also Figures 1 to 6 , the present invention proposes a tea tree pest dynamic detection system, the system includes an observation device, a control system and an improved tea tree pest detection model;

[0045] Among them, it should be noted that the observation equipment is used for data processing and coordination, capturing images of tea garden insect conditions while achieving high-speed and stable data transmission. The control system provides an interactive operation interface for remote observation equipment for server connection, equipment control and image data upload. The improved model for tea tree pest detection is based on the Yolov8 network, and the Faster Block module is used to replace the C2f module. At the same time, the P5 target detection layer of the PAN-FPN structure in the original neck network is removed, and the P2 small target detection layer is added to the upper part of the structure to identify and locate the pest position. The improved model is defined as YOLOv8-FasterTea.

[0046] It should also be noted that the observation equipment also includes a microcomputer, a color USB high-definition industrial camera, a 5G outdoor WIFI module, an energy system and a power supply voltage regulation conversion module:

[0047] Among them, the microcomputer is a Raspberry Pi 4B microcomputer, which is used for data processing and coordination. The resolution of the color USB high-definition industrial camera is 12 million pixels, which is used to capture insect images in the tea garden and is used for data transmission. The energy system and power supply voltage stabilization conversion module. The energy system consists of a 120W solar panel and a 240AH battery, which is used to provide energy support for observation equipment in outdoor environments. The power supply voltage stabilization conversion module is used for stable power supply under different voltage conditions.

[0048] It should also be noted that the control system uses the Vue technology framework to write the control interface, and the operating logic is implemented in Python. The control system includes a server, a device control unit, and an image data upload module. The server uses FRP technology to implement the web interface to control the remote control terminal (i.e., Raspberry Pi) through a pre-configured IP address. The device control unit uses the OpenCV library to control the camera, set the timed photo function, and reserve the light on and camera movement and steering functions to transmit the captured insect pest images to the remote image storage server. The image information is forwarded by the server to other servers for storage. After obtaining the token of the image storage server, the image data is transmitted according to the specified API format.

[0049] It should also be noted that the improved tea tree pest detection model is used for image feature extraction, and includes the following steps:

[0050] S1. Create a partial convolution operator for spatial feature extraction. Apply conventional convolution to a portion of the input channels for spatial feature extraction, and keep the remaining channels unchanged, as shown in Equation (1) and Equation (2):

[0051] FLOPs pconv =h·w·k 2 ·cp 2 (1);

[0052] Variable explanation:

[0053] h: The height of the input feature map.

[0054] w: width of the input feature map.

[0055] k: The size of the convolution kernel.

[0056] c p : Number of input channels (Input Channels).

[0057] FLOPs refers to the number of floating point operations, which is an important indicator to measure the computational complexity of an algorithm. In the convolution operation, each multiplication and addition is counted as a floating point operation.

[0058] For standard convolution operations, FLOPs are calculated based on the multiplication and summation of each position of the input feature map with the convolution kernel.

[0059] h·w in the formula represents the number of spatial locations of the input feature map, k 2 Represents the number of elements in the convolution kernel, c p 2 Represents the interaction between input channels (because some convolution operations may involve complex calculations between channels).

[0060] MAC pconv =h·w·2c p +k 2 ·c p 2 ≈h·w·2c p (2);

[0061] Variable explanation:

[0062] Same meaning as the variables in the above FLOPs formula;

[0063] MAC meaning;

[0064] MAC refers to the memory access cost, which is the number of times data is read from and written to the memory during the calculation process. In the convolution operation, MAC is an indicator of the memory overhead of the model at runtime, and pconv is a partial convolution.

[0065] h·w·2c in the formula p Represents the memory access cost of the input feature map and the output feature map (because each position needs to read and write data, considering the input and output channels, it is multiplied by 2c p ).

[0066] k 2 ·c p 2 represents the memory access cost associated with the convolution kernel operation. Since in practice, k 2 ·c p 2 Usually more than h.w.2c p Much smaller (especially for larger input feature maps), so MAC can be considered pconv ≈h·w·2c p .

[0067] FasterNet Block sets two PWConv layers after the PConv layer and only places normalization and activation layers after the intermediate layers;

[0068] S2. Remove the P5 target detection layer of the PAN-FPN structure in the original neck network, and add a P2 small target detection layer on the upper part of the structure. The P2 small target detection layer is used to perform feature fusion on the feature maps of the first few layers in the backbone network.

[0069] Example

[0070] In practical applications, the above system specifically includes the following architecture and implementation methods:

[0071] As an innovative tool designed specifically for outdoor tea garden environments, the Tea Tree Pest Monitoring System is committed to helping tea growers quickly detect and respond to pest challenges in tea gardens, in order to achieve the goal of improving tea production and quality;

[0072] The system integrates cutting-edge IoT technology, deep learning algorithms, and deep agricultural expertise to provide tea garden managers with a comprehensive pest monitoring and early warning solution;

[0073] See also Figure 1 and Figure 2 A detailed description of the system architecture. In pursuit of environmental protection and energy saving, the system cleverly uses solar panels with voltage conversion modules to achieve energy self-sufficiency while maintaining hardware security.

[0074] Raspberry Pi 4B, as the core of data processing, is connected to the Internet through 5G outdoor WIFI to ensure the regular upload of monitoring data, providing solid data support for subsequent data analysis and model training;

[0075] In addition, the system places special emphasis on the lightweight design of the model, which facilitates its subsequent deployment directly on the data collection nodes;

[0076] 1. Tea tree insect monitoring device:

[0077] The tea tree insect monitoring device consists of the following core components:

[0078] A Raspberry Pi 4B microcomputer serves as the brain of the system, responsible for data processing and coordination;

[0079] A color USB high-definition industrial camera with a resolution of up to 12 megapixels to capture images of insects in tea gardens;

[0080] A high-efficiency energy system consisting of a 120W solar panel and a 240AH battery ensures continuous operation of the device in outdoor environments;

[0081] A precise power supply voltage conversion module ensures stable power supply of the system under different voltage conditions;

[0082] A 5G outdoor WIFI module for data transmission;

[0083] All components of the system are installed on a solid steel structure of poles and cross arms to adapt to the challenges of outdoor environments and ensure the long-term stable operation of the monitoring device;

[0084] The tea tree insect monitoring device is installed on a 3-meter-high pole. A sleeve is set in the middle of the pole to support the cross arm for installing the monitoring camera and the insect observation steel plate. The high-definition camera is placed 50 cm away from the insect observation plate. In order to improve the neatness of the line layout and the operational flexibility of the equipment, the outlet holes at the bottom and sides and the universal joints installed on each cross arm are provided.

[0085] In addition, the equipment is stabilized by a solid steel cage and heavy flanges, ensuring that the device can operate safely and stably even in bad weather;

[0086] All key control equipment, including Raspberry Pi 4B and power supply voltage conversion modules, are safely encapsulated in the stainless steel distribution box of the pole to achieve good electromagnetic isolation and protection;

[0087] After a laboratory testing phase, the tea tree insect monitoring system was deployed in a tea garden in Sichuan Province, China;

[0088] Monitoring devices were carefully placed at several key locations in the tea garden to collect data 24 hours a day. The entire monitoring network covers a vast area of ​​25,000 square meters. A total of three sets of monitoring devices were deployed, each of which is responsible for monitoring an area of ​​approximately 20 square meters;

[0089] The installation location of the device is selected based on the living habits of pests according to the professional advice of tea experts, and is a hot spot where pest activities are more concentrated;

[0090] 2. Outdoor insect observation equipment control system:

[0091] See also Figure 3 ,The control interface of this system is written in Vue technology framework, and the operation logic is implemented in Python language;

[0092] The development environment of this system is shown in Table 1:

[0093] Table 1 Development environment of this system

[0094] operating system Windows 10 (x64) Develop software Pycharm\Vscode Industrial Camera Ultra-light images Outdoor Control / Data Unit Raspberry Pi 4B

[0095] The following is a detailed description of the technical solutions for each module:

[0096] 1. Server connection:

[0097] FRP technology (Fast Reverse Proxy) is a network service that acts as an intermediary between the server and the client, forwarding the client's request to the backend server and receiving the server's response before forwarding it to the client;

[0098] The FRP workflow mainly includes the following aspects:

[0099] 1) Proxy request:

[0100] The client does not communicate directly with the target server, but sends the request to the FRP proxy server;

[0101] 2) Request forwarding:

[0102] After receiving the client's request, FRP forwards the request to the corresponding backend server according to the configured rules;

[0103] 3) Server processing:

[0104] After receiving the forwarded request, the backend server processes it and generates response data;

[0105] 4) Response forwarding:

[0106] The backend server sends the processing result back to the FRP proxy server, and FRP forwards the response to the client of the original request;

[0107] Therefore, through this technology, I can control the remote control terminal (that is, the Raspberry Pi connected to the network in the device) through the pre-configured IP address on the web interface.

[0108] 2. Device control - camera control:

[0109] Use the open source OpenCV library to control the industrial camera connected to the Raspberry Pi and set the timed photo function. Here I also reserve the light on and camera movement and steering functions (night monitoring + adjusting the photo position);

[0110] At the same time, the insect pest images taken will be transmitted to the remote image server, and the image information will be forwarded by the MQTT server to other servers for storage;

[0111] Core code:

[0112]

[0113]

[0114]

[0115]

[0116]

[0117]

[0118] First, obtain the token (authorization) of the image storage server and transmit the image data according to the specified API format;

[0119] 3. Outdoor insect detection algorithm:

[0120] In order to improve the detection accuracy of small flying insects in images and ensure the lightweight of the model, this patent proposes an improved model specifically for tea tree pest detection based on the Yolov8 network, named Yolov8-FasterTea;

[0121] Based on the Yolov8 algorithm, this model makes targeted adjustments and optimizations to more effectively identify and locate pest locations;

[0122] See also Figure 4 and Figure 5 The outdoor insect detection algorithm not only improves detection accuracy, but also significantly reduces the model size compared to before, providing a more accurate and lightweight solution for monitoring and prevention of tea tree pests;

[0123] As can be seen from the figure, YOLOv8-FasterTea replaces the original C2f module of the YOLOv8 network with a lighter Faster Block (FBlock) module;

[0124] The traditional C2f structure connects multiple Bottlenecks in series, adds more skip-layer connections, cancels the convolution operation in the branch, and adds additional split operations to enrich the feature information extraction;

[0125] However, this structure makes the computational complexity and parameter quantity large, and has certain limitations for the extraction of small target features;

[0126] To address this problem, this paper partially replaces the module by using the FBlock module;

[0127] The specific module introduction will be described in detail in the following sections;

[0128] At the same time, considering the actual application scenario of pest detection in tea gardens, this paper chooses to add the P2 small target layer while removing the original P5 detection layer in YOLOv8;

[0129] This targeted improvement strategy not only optimizes the resource allocation of the model, but also significantly improves the accuracy and efficiency of small target flying insect detection;

[0130] Through the P2 small target layer, the model can more sensitively capture subtle signs of pests in tea gardens, which is crucial for early detection and prevention of pests.

[0131] At the same time, removing the P5 detection layer not only simplifies the network structure, but also reduces the computational burden of the model, making the model run faster while maintaining high accuracy, making it more suitable for deployment in resource-constrained field environments;

[0132] Such optimization makes the model more flexible and efficient in practical applications, and can quickly respond to the needs of tea garden pest detection;

[0133] Specifically, the improved tea tree pest detection model is used for image feature extraction, including the following steps:

[0134] Create a partial convolution operator for spatial feature extraction, apply regular convolution on a portion of the input channels for spatial feature extraction, and keep the remaining channels unchanged, see equations (1) and (2):

[0135] FLOPs pconv =h·w·k 2 ·c p 2 (1);

[0136] MAC pconv =h·w·2c p +k 2 ·c p 2 ≈h·w·2c p (2);

[0137] FasterNet Block sets two PWConv layers after the PConv layer and only places normalization and activation layers after the intermediate layers;

[0138] Among them, the model structure design yaml file is:

[0139]

[0140]

[0141]

[0142] The FasterNet Block structure is implemented as:

[0143]

[0144]

[0145]

[0146] Through the above scheme, the present invention adopts an ultra-high-definition microscopic camera with excellent observation ability; the operation process is simple, the user can get started quickly, support remote deployment and debugging, save time and cost, the original algorithm is efficient and energy-saving, and is suitable for a variety of deployment environments; the sticky insect trapping technology is economical and practical, and easy to maintain;

[0147] In addition, the equipment has low cost, self-power supply and networking, easy installation and operation, and plug-and-play, providing tea gardens with a comprehensive set of pest monitoring and early warning solutions, significantly improving tea production and quality.

[0148] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Tea tree pest dynamic detection system, characterized in that: Including observation equipment, control system and tea tree pest detection model; The observation equipment is used for data processing and coordination, capturing images of tea garden insect conditions and achieving high-speed and stable data transmission; The control system provides an interactive operation interface for remote observation equipment, which is used for server connection, equipment control and image data uploading; The improved tea tree pest detection model is based on the Yolov8 network, using the Faster Block module to replace the C2f module, while removing the P5 target detection layer of the PAN-FPN structure in the original neck network, adding a P2 small target detection layer on the upper part of the structure to identify and locate the pest position, and the improved model is defined as YOLOv8-FasterTea.

2. The tea tree pest dynamic detection system according to claim 1, characterized in that: The observation device also includes: microcomputers for data processing and coordination; Color USB HD industrial camera to capture images of insects in tea gardens; 5G outdoor WIFI module for data transmission.

3. The tea tree pest dynamic detection system according to claim 2, characterized in that: The observation equipment also includes an energy system and a power supply voltage stabilization conversion module, which are used to provide energy support for the observation equipment in an outdoor environment. The power supply voltage stabilization conversion module is used to provide stable power supply under different voltage conditions.

4. The tea tree pest dynamic detection system according to claim 1, characterized in that: The control system adopts Vue technology framework to write the control interface, and the operation logic is implemented in Python language.

5. The tea tree pest dynamic detection system according to claim 4, characterized in that: The control system includes a server, a device control unit and a picture data uploading module; The server uses FRP technology to realize the control of the remote control terminal through the pre-configured IP address by the web page interface; The device control unit uses the OpenCV library to control the camera, set the timed photo function, and reserve the light on and camera movement functions to transmit the captured insect pest images to the remote image bed server, and the image information is forwarded by the server to other servers for storage; After obtaining the token of the image storage server, the image data is transmitted according to the specified API format.

6. The tea tree pest dynamic detection system according to claim 1, characterized in that: The improved tea tree pest detection model is used for image feature extraction, and comprises the following steps: S1. Create a partial convolution operator for spatial feature extraction, apply convolution on part of the input channels for spatial feature extraction, and keep the remaining channels unchanged, see equations (1) and (2): FLOPs pconv =h·w·k 2 ·c p 2 (1); MOTHER pconv =h w 2c p +k 2 c p 2 ≈h w 2c p (2); In formula (1) and formula (2), h is the height of the input feature map, w is the width of the input feature map, k is the size of the convolution kernel, and c is p is the number of input channels; In formula (1) and formula (2), FLOPs is the amount of computation; MAC is the memory access cost, and pconv is partial convolution; FasterNet Block sets two PWConv layers after the PConv layer and only places normalization and activation layers after the intermediate layers; S2. Remove the P5 target detection layer of the PAN-FPN structure in the original neck network, and add a P2 small target detection layer on the upper part of the structure. The P2 small target detection layer is used to perform feature fusion on the feature maps of the first few layers in the backbone network.

Images