Method for detecting plant diseases and insect pests of open field vegetables

Abstract

The invention discloses an open field vegetable disease and insect pest detection method, and relates to the technical field of disease and insect pest detection methods. A multi-level data acquisition mechanism is designed, multi-spectral unmanned aerial vehicle high-altitude aerial photography and ground intelligent inspection vehicle fine image sampling are combined for the first time, a double-level and multi-view-angle data acquisition system is formed, and the spatial resolution and detection accuracy of disease and pest area recognition are improved; secondly, a YOLOv11 model structure is improved, and a standard convolution structure in a YOLOv11 backbone network is replaced with a PConv lightweight module, so that the model parameter quantity is reduced, the reasoning speed is increased, and the method adapts to an edge device deployment scene; besides, automatic generation of a visual disease distribution diagram and a pesticide application operation diagram is realized, a detection result is combined with positioning information, the disease and pest distribution diagram and the pesticide application operation diagram are automatically generated, an accurate and scientific basis is provided for agricultural pesticide application operation, and the intelligent level of field operation is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of pest and disease detection methods, and in particular relates to a method for detecting pest and disease on open-field vegetables. Background Art

[0002] As an important part of agricultural production, open-field vegetables are susceptible to pests and diseases during their growth process, resulting in reduced yields, deteriorating quality, and even regional agricultural economic losses. Traditional pest and disease detection mainly relies on manual field observation and empirical judgment of abnormal crop symptoms. However, this method is inefficient and highly subjective, and it is difficult to detect early diseases or hidden pests in a timely manner. Misjudgments are prone to cause delayed prevention and control. In recent years, intelligent detection technology based on image recognition has gradually been applied. However, due to the limitations of natural light changes, occlusion of branches and leaves, and complex background interference, the generalization and robustness of existing algorithms in open-air environments are insufficient, especially for small sample diseases or morphologically variant insects. The recognition accuracy is significantly reduced. In addition, deep learning models rely on large-scale labeled data training, which has high implementation costs and is difficult to adapt to the diversity characteristics of regional pests and diseases.

[0003] The existing technology mainly has the following disadvantages:

[0004] (1) Unable to achieve full-process unmanned detection: The image acquisition and decision-making in the existing plant disease and insect pest technology links rely on manual operation. For example, the decision-making of the Chinese invention patent CN117333862A "A method, device and system for monitoring the growth status of green vegetables" requires manual shooting and collection of image information, and then experts remotely analyze and give guidance. This has many links, a long cycle and low decision-making efficiency.

[0005] (2) Existing technologies have a single method for acquiring data: Existing technologies often use a single device to acquire data;

[0006] (3) The existing technology lacks real-time processing capabilities and has a low level of intelligence: For example, after the model is constructed in the Chinese invention patent CN117333862A “A method, device and system for monitoring the growth of green vegetables”, experts need to make remote judgments and return the expert opinions to farmers. Decisions need to be made manually and real-time decisions cannot be made;

[0007] (4) The existing technology lacks pest and disease treatment strategies and cannot achieve closed-loop control from detection to prevention and control: for example, the Chinese invention patent CN119723323A, a rice pest and disease detection method and system, only detects the type of pest and disease.

[0008] Therefore, in view of the above problems, the present invention proposes a method for detecting plant diseases and insect pests in open fields. Summary of the Invention

[0009] (1) Technical problems to be solved by the present invention:

[0010] The purpose of the present invention is to provide a method for detecting plant diseases and insect pests in open fields to solve the problems raised in the background technology.

[0011] (2) In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions:

[0012] A method for detecting plant diseases and insect pests in open fields, comprising the following steps:

[0013] S1. Use a multispectral drone to take aerial photos of the entire experimental area to obtain a multispectral image of the experimental area;

[0014] S2. Stitching the obtained multispectral images, analyzing the health of the vegetables based on visible light and NDVI index, identifying areas of abnormal growth, and determining them as suspected areas of pest and disease occurrence;

[0015] S3. Complete the inspection vehicle’s path planning based on the experimental area’s path distribution;

[0016] S4. After completing the path planning, use the inspection vehicle to sample pest and disease images from multiple directions and angles along the planned path. When the inspection vehicle arrives at the abnormal point, it takes RGB images of the locations where vegetable pests and diseases occur in the experimental area and uses the improved YOLOv11 detection model to detect the captured pest and disease images.

[0017] S5. After the test is completed, the test results are combined with the geographical location information to record the location of each pest and disease occurrence and generate a pest and disease distribution map;

[0018] S6. Generate a pesticide application map based on the obtained pest and disease distribution map, and use the pesticide application map to implement differentiated and precise pesticide application operations for the pest and disease areas using pesticide application equipment.

[0019] Preferably, the S3 specifically includes the following contents:

[0020] S3.1. Plan the optimal route from the inspection vehicle's starting position to the first abnormal point;

[0021] S3.2. Combine the nearest neighbor strategy to connect the remaining abnormal points at once to implement inspection vehicle path planning to ensure the shortest driving path.

[0022] Preferably, the improved YOLOv11 detection model described in S4 specifically includes the following contents:

[0023] Use PConv to replace all ordinary convolution modules except the first two ordinary convolution layers in the Backbone layer of the YOLOv11 model to improve the model's inference speed and facilitate the model's deployment on embedded devices;

[0024] A small target detection head is added. A repeated "upsampling-cascade-C3k2" three-layer structure is added after the second C3k2 module in the Head layer of the original network structure, and then a "convolution-cascade-C3k2" three-layer structure is added. Here, the output of C3k2 is used as the input of the model detection head, and the input of the detection head is increased from the original 3 C3k2 modules to 4.

[0025] (3) The beneficial effects of the present invention include the following:

[0026] (1) Multi-dimensional data acquisition: The present invention obtains data information through two data acquisition methods: drone aerial photography and inspection vehicle sampling, ensuring the accuracy of map location distribution and laying the foundation for accurate disease data collection.

[0027] (2) Application guidance function with stronger operability: After detecting pests and diseases, the present invention generates a disease distribution map and a pesticide application map according to the type and location of the disease, and has the ability to coordinate agricultural machinery operations, providing decision support for agricultural pesticide application.

[0028] (3) Balancing model lightweighting and small target optimization: This paper introduces the PConv lightweight module and small target detection branch into YOLOv11, which greatly improves the running speed of the model without sacrificing detection accuracy. It is suitable for resource-constrained edge device deployment scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a flow chart of a method for detecting plant diseases and insect pests in open fields proposed by the present invention;

[0030] Figure 2 The multispectral image after stitching proposed in Example 1 of the present invention;

[0031] Figure 3 Schematic diagram of the improved YOLOv11 pest and disease detection model proposed in Example 1 of the present invention;

[0032] Figure 4 This is a structural diagram of the PConv module proposed in Example 1 of the present invention;

[0033] Figure 5 This is a schematic diagram of the pest and disease detection results proposed in Example 1 of the present invention;

[0034] Figure 6 This is a diagram of the pesticide application operation proposed in Example 1 of the present invention. DETAILED DESCRIPTION

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0036] The present invention proposes a method for detecting pests and diseases in open-field vegetables, which implements a complete process from generating anomaly distribution maps using drone imagery, obtaining specific pest and disease images through inspection vehicles, and finally generating pesticide application maps after pest and disease detection. Compared to existing technologies, the present invention designs a multi-level data acquisition mechanism, combining multispectral drone aerial photography with fine image sampling by intelligent ground inspection vehicles for the first time to form a dual-level, multi-perspective data acquisition system, improving the spatial resolution and detection accuracy of pest and disease area identification. Secondly, the present invention improves the YOLOv11 model structure by replacing the standard convolutional structure in the YOLOv11 backbone network with a lightweight PConv module, reducing the number of model parameters, improving inference speed, and adapting to edge device deployment scenarios. Furthermore, the present invention realizes the automatic generation of visual pest distribution maps and pesticide application operation maps, combining detection results with positioning information to automatically generate pest and disease distribution maps and pesticide application operation maps, providing an accurate and scientific basis for agricultural pesticide application operations and improving the intelligent level of field operations. The following describes a method for detecting outdoor vegetable pests and diseases in accordance with the present invention in conjunction with the accompanying drawings and specific examples. The specific details are as follows.

[0037] Example 1:

[0038] The present invention proposes a method for detecting field vegetable pests and diseases, which mainly includes a data acquisition platform and a data processing module. Figure 1 As shown, it specifically includes the following contents.

[0039] Step 1:

[0040] Use the multispectral drone DJI Mavic 3M to take aerial photos of the entire farmland area under monitoring;

[0041] Step 2:

[0042] The obtained multispectral images are stitched together and the health of vegetables is analyzed based on visible light and NDVI index. Crops infected by pests and diseases generally show yellowing and necrosis of leaves, reduced leaf area, etc., which are manifested in spectral performance as decreased green light absorption, decreased red light absorption, decreased NIR reflectance, etc., and reflected in the NDVI value as a decrease in NDVI value. Therefore, the NDVI value can be used to reflect the presence of pests and diseases to a certain extent. Identify areas of abnormal growth and determine them as suspected areas of pest and disease occurrence (i.e., abnormal point areas);

[0043] UAV aerial images can be ortho-mosaiced using professional software such as Pix4d Mapper to obtain a base map of the crop area. An example of a multispectral image after stitching is shown below. Figure 2 shown.

[0044] Using the near-infrared band image and the red band image obtained after ortho-mosaic processing, the NDVI index is calculated according to the following formula to obtain the NDVI base map of the working area:

[0045]

[0046] Where, NIR represents the near-infrared band image; Red represents the red light band image;

[0047] The OpenCV image processing library is used to identify the NDVI base map. By analyzing the color features in the base map and monitoring areas with low NDVI values, it is possible to determine the areas where suspected crop pests and diseases are concentrated.

[0048] Step 3:

[0049] The inspection vehicle's path planning is completed based on the distribution of farmland paths. First, the optimal path is planned based on the inspection vehicle's starting position to the location of the first outlier point. Then, the nearest neighbor strategy is used to connect the remaining outliers at a time to achieve the inspection vehicle's path planning and ensure the shortest driving path.

[0050] Step 4:

[0051] After completing the path planning, the inspection vehicle is used to collect pest and disease images from multiple directions and angles along the planned path. After the inspection vehicle arrives at the abnormal point, RGB images are taken of the location where the vegetable pests and diseases occur. The detection model is used on the pest and disease images, and the improved YOLOv11 model is used for detection. The improved model is shown in the figure below. Figure 3 As shown:

[0052] The following improvements are made to the YOLOv11 model. First, PConv is used to replace the other ordinary convolution modules in the Backbone layer except the first two ordinary convolution layers. The module structure diagram of PConv is as follows: Figure 4 The PConv module is a lightweight convolution module that can improve model speed. Using it to replace the original ordinary convolution module can improve the model's inference speed and facilitate the model's deployment on embedded devices.

[0053] Secondly, because the targets of detection are pests and diseases, which are typically small, a small target detection head was added to improve the model's detection capabilities. Specifically, a repeated three-layer structure of "upsampling-cascade-C3k2" was added after the second C3k2 module in the Head layer of the original network structure, followed by a three-layer structure of "convolution-cascade-C3k2". The output of this C3k2 serves as the input to the model detection head, increasing the number of C3k2 inputs from the original three to four.

[0054] The improved YOLOv11 model can be used to detect pests and diseases in captured images. The detection results are as follows: Figure 5 As shown in the figure, the model also classifies the types of diseases detected, and the points where the diseases and pests occur are marked with boxes.

[0055] Step 5:

[0056] After the test is completed, the test results are combined with the geographical location information to record the location of each pest and disease occurrence and generate a pest and disease distribution map;

[0057] Step 6:

[0058] Generate a pesticide application map based on the disease distribution map, including the application plan, dosage and corresponding application parameters, to guide the application equipment to carry out differentiated and precise pesticide application in the pest and disease area. Figure 6 , uniformly apply pesticides to all areas, and carry out differentiated pesticide application operations targeting specific pests and diseases in areas where pests and diseases exist. The pesticide application path planning method is the same as the above-mentioned inspection vehicle path planning method.

[0059] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes based on the technical solutions and improved concepts of the present invention within the technical scope disclosed by the present invention, and these changes should be covered by the scope of protection of the present invention.

Claims

1. A method for detecting plant diseases and insect pests in open fields, characterized in that: The steps include: S1. Use a multispectral drone to take aerial photos of the entire experimental area to obtain a multispectral image of the experimental area; S2. Stitching the obtained multispectral images, analyzing the health of the vegetables based on visible light and NDVI index, identifying areas of abnormal growth, and determining them as suspected areas of pest and disease occurrence; S3. Complete the inspection vehicle’s path planning based on the experimental area’s path distribution; S4. After completing the path planning, use the inspection vehicle to sample pest and disease images from multiple directions and angles along the planned path. When the inspection vehicle arrives at the abnormal point, it takes RGB images of the locations where vegetable pests and diseases occur in the experimental area and uses the improved YOLOv11 detection model to detect the captured pest and disease images. S5. After the test is completed, the test results are combined with the geographical location information to record the location of each pest and disease occurrence and generate a pest and disease distribution map; S6. Generate a pesticide application map based on the obtained pest and disease distribution map, and use the pesticide application map to implement differentiated and precise pesticide application operations for the pest and disease areas using pesticide application equipment.

2. A method for detecting field vegetable pests and diseases according to claim 1, characterized in that: The S3 specifically includes the following contents: S3.

1. Plan the optimal route from the inspection vehicle's starting position to the first abnormal point; S3.

2. Combine the nearest neighbor strategy to connect the remaining abnormal points at once to implement inspection vehicle path planning to ensure the shortest driving path.

3. A method for detecting outdoor vegetable pests and diseases according to claim 1, characterized in that: The improved YOLOv11 detection model described in S4 specifically includes the following: Use PConv to replace all ordinary convolution modules except the first two ordinary convolution layers in the Backbone layer of the YOLOv11 model to improve the model's inference speed and facilitate the model's deployment on embedded devices; A small object detection head is added. After the second C3k2 module in the Head layer of the original network structure, a repeated "upsampling-cascade-C3k2" three-layer structure is added, and then a "convolution-cascade-C3k2" three-layer structure is added. Here, the output of C3k2 is used as the input of the model detection head, and the input of the detection head is increased from the original three C3k2 modules to four.

Images