A method, apparatus, device, and storage medium for identifying grassland rodent burrows.

Abstract

This invention discloses a method, apparatus, device, and storage medium for identifying mouse holes in grasslands, relating to the field of mouse hole identification technology. The method includes the following steps: acquiring multiple historical UAV images of a grassland area; determining the target scale of the mouse hole to be identified based on the bounding boxes of mouse holes in the multiple historical UAV images; constructing an improved YOLOv8n model based on the target scale; acquiring images of the UAV to be identified in the grassland area; cropping the images and inputting them into the improved YOLOv8n model to obtain the corresponding mouse hole identification result. The operation of this invention is simple, its mouse hole identification accuracy does not depend on the interpreter's experience, and the improved YOLOv8n model has a simple structure, greatly improving the real-time performance of grassland mouse hole identification.

Description

Technical Field

[0001] This invention relates to the field of rodent burrow identification technology, and in particular to a method, apparatus, device, and storage medium for identifying rodent burrows in grasslands. Background Technology

[0002] Currently, the identification of rodent burrows in grasslands in my country mainly relies on manual field surveys. While these methods are simple and intuitive to operate and require low technical skills from personnel, they suffer from the following problems: First, the results of rodent infestation surveys are related to the number and representativeness of the selected quadrats. Too many quadrats result in high manpower and economic costs, and quadrat selection is limited by geographical conditions, leading to poor accuracy of the survey results. Second, traditional survey methods focus on the rodent population itself, are complex to operate, lack real-time performance, and cannot be used for repeated and large-scale monitoring of grassland rodent infestation areas. When large-scale rodent infestations occur, it can easily lead to untimely reporting of rodent infestation information.

[0003] Unmanned aerial vehicle (UAV) systems integrate technologies such as unmanned aerial vehicles, GPS navigation and positioning, and aerial remote sensing, offering advantages such as high spatial resolution, timely information acquisition, low cost, and minimal cloud interference. Target detection based on UAV imagery is widely used in emergency disaster relief, power line inspection, and traffic monitoring. Researchers can intelligently analyze images captured by UAVs to quickly and efficiently extract ground feature information, saving significant manpower and resources. Plot-scale UAV remote sensing has become one of the main forms of aerial remote sensing, providing a novel method for identifying grassland rodent burrows.

[0004] Existing UAV image-based rodent hole identification technologies fall into two categories: visual interpretation and object-oriented image analysis (OBIA). The former utilizes manual visual interpretation to statistically analyze rodent holes in visible light images captured by UAVs, performing GIS grid and GIS overlay analysis to obtain spatial distribution trends and coverage rates of rodent hole clusters; this is a semi-automated rodent hole identification technique. OBIA technology can obtain homogeneous object information and is becoming increasingly common in high-resolution UAV image analysis. Current rodent hole identification research combines OBIA with machine learning techniques such as template matching, random forests, and decision trees, achieving an accuracy rate of 87.17%. These studies demonstrate that using UAV technology for rodent hole identification has significant practical application value, providing an efficient and intelligent solution for grassland rodent control.

[0005] Visual interpretation of mouse holes based on UAV images involves a complex process, and the accuracy of mouse hole identification depends heavily on the interpreter's experience, making it impractical. The extended feature space of homogeneous objects in the OBIA method, such as spectral, texture, and shape features, significantly increases the complexity of image analysis, resulting in longer processing times and reduced real-time performance. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the present invention provides a grassland mouse hole identification method, device, equipment and storage medium, which solves the problems of complex operation process of existing mouse hole identification based on visual interpretation of UAV images and long data processing time of the OBIA method.

[0007] The present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a method for identifying grassland rodent burrows, comprising the following steps:

[0009] Acquire multiple historical drone images of the grassland area;

[0010] The target scale of the mouse hole to be identified is determined by the target bounding box of the mouse hole in multiple historical UAV images. Based on the target scale, the target detection head and the target feature integration module in the neck of the original YOLOv8n model are determined.

[0011] The remaining detection heads other than the target detection head in the head network of the original YOLOv8n model are removed, and the remaining feature integration modules other than the target feature integration module in the neck network are removed. The structure in which the third C2f layer and SPPF layer in the backbone network of the original YOLOv8n model are connected to the second C2f layer in the neck network through the Upsample module and the Concat module is retained to obtain the improved YOLOv8n model.

[0012] Images of drones to be identified are collected in the grassland area. After cropping the images, they are input into the improved YOLOv8n model to obtain the corresponding mouse hole identification results.

[0013] Preferably, determining the target scale of the mouse hole to be identified based on the target bounding box of the mouse hole in multiple historical UAV images includes the following steps:

[0014] Each historical drone image is cropped to obtain multiple cropped images. Mouse holes in the cropped images are labeled to obtain multiple bounding boxes.

[0015] The size of multiple target bounding boxes is statistically analyzed to obtain the target scale of the mouse hole to be identified.

[0016] Preferably, the historical drone image has a resolution of 5280 pixels × 3956 pixels, the multiple cropped images have a resolution of 300 pixels × 300 pixels, the width and height of the multiple target boxes are distributed within a region of 0.1 × 0.1, and the target size of the mouse hole to be identified is within 30 × 30 pixels.

[0017] Preferably, the remaining detection heads after removing the target detection heads from the head network of the original YOLOv8n model include 20×20 and 40×40 detection heads.

[0018] Preferably, before sequentially inputting multiple images to be recognized into the improved YOLOv8n model, the improved YOLOv8n model needs to be trained. The training process includes the following steps:

[0019] Multiple historical drone images are filtered, and the filtered historical drone images are cropped to obtain multiple sample images, each of which has mouse holes.

[0020] The mouse holes in multiple sample images are labeled to obtain the dataset, which is then divided into training, testing and validation sets in an 8:1:1 ratio.

[0021] The improved YOLOv8n model is trained using the training set, and tested and validated using the test and validation sets.

[0022] Secondly, the present invention provides a grassland rodent burrow identification device, comprising:

[0023] The acquisition module is used to acquire multiple historical drone images of the grassland area;

[0024] The determination module is used to determine the target scale of the mouse hole to be identified based on the target bounding box of the mouse hole in multiple historical UAV images, and to determine the target detection head and neck feature integration module in the head network of the original YOLOv8n model based on the target scale.

[0025] The building module is used to remove the remaining detection heads other than the target detection heads in the head network of the original YOLOv8n model, remove the remaining feature integration modules other than the target feature integration module in the neck network, and retain the structure in the backbone network of the original YOLOv8n model where the third C2f layer and SPPF layer are connected to the second C2f layer in the neck network through the Upsample module and the Concat module, thus obtaining the improved YOLOv8n model;

[0026] The recognition module is used to collect images of drones to be identified in the grassland area. After cropping the images, the drone images are input into the improved YOLOv8n model to obtain the corresponding mouse hole recognition results.

[0027] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described grassland rodent burrow identification method.

[0028] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described grassland rodent burrow identification method.

[0029] Compared with the prior art, the above-mentioned at least one technical solution adopted by the present invention can achieve the following beneficial effects:

[0030] This invention first determines the target scale based on the bounding boxes of mouse holes in historical UAV images of grassland areas. Based on this target scale, the original YOLOv8n model is improved. Specifically, the remaining detection heads in the head network of the original YOLOv8n model (excluding the target detection head) and the remaining feature integration modules in the neck network (excluding the target feature integration module) are removed, reducing network complexity. Simultaneously, the structure of the third C2f layer and SPPF layer in the backbone network of the original YOLOv8n model, connected to the second C2f layer in the neck network through Upsample and Concat modules, is retained. This invention preserves the fusion of deep semantic information and shallow geometric information in the backbone network, effectively extracting features for mouse hole recognition and significantly improving recognition accuracy. The operation of this invention is simple, and its mouse hole recognition accuracy does not depend on the interpreter's experience. Furthermore, the improved YOLOv8n model has a simple structure, greatly improving the real-time performance of grassland mouse hole recognition. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the UAV image acquisition hardware system of the present invention;

[0033] Figure 2 Flowchart for constructing the dataset of this invention;

[0034] Figure 3 This is a schematic diagram of the original YOLOv8n model structure;

[0035] Figure 4 This is a relative scale statistical diagram of the mouse hole data prior box of the present invention;

[0036] Figure 5 This is a schematic diagram of the improved YOLOv8n model structure of the present invention;

[0037] Figure 6The image shows the detection results of the original YOLOv8n model and the improved YOLOv8n model of this invention for grassland mouse burrows;

[0038] in, Figure 6 (a): Detection results of the original YOLOv8n model for grassland mouse burrows. (b): Detection results of the improved YOLOv8n model for grassland mouse burrows. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] To address the shortcomings of existing technologies, this invention provides a grassland mouse burrow identification method that fully considers the relationship between the UAV's flight altitude and the target's prior bounding box, further fusing different levels of target features to improve identification accuracy and speed. The method specifically includes the following steps:

[0041] S1: Acquire multiple historical drone images of the grassland area.

[0042] Reference Figure 1 The UAV image acquisition and preprocessing consists of a UAV system and an image processing system. The UAV system mainly comprises the UAV, sensors, remote controller, data transmission system, and image processing system. The image processing system primarily uses flight path planning software and image stitching software. Flight path planning utilizes the automatic flight control ground station software Altizure, which allows for free setting of parameters such as flight area, altitude, and overlap rate. Generally, the forward overlap rate is set to 65%, and the lateral overlap rate is set to 35%. Due to the windy weather in the test area, to avoid image loss and affecting post-processing, both the forward and lateral overlap rates are set to 80% in this invention.

[0043] In this embodiment, a DJI M210 RTK V2 quadcopter drone equipped with a Zenmuse X5S RGB camera is used for data acquisition.

[0044] Reference Figure 2In March 2020 and April 2021, drone images were captured in a grassland area under clear, windless weather conditions. Images were first filtered to remove those blurry due to camera exposure and drone shaking. The image resolution was 5280 pixels × 3956 pixels. The remaining images were cropped to 300 pixels × 300 pixels and then filtered again, retaining images with mouse holes and removing those without, resulting in 1076 samples. Rectangular bounding boxes were used in the Labellimg software to label the mouse holes in the images, ensuring the edges of the bounding boxes were tangent to the edges of the mouse holes to guarantee accuracy. Finally, the dataset was divided into training, testing, and validation sets in an 8:1:1 ratio.

[0045] S2: Determine the target scale of the mouse hole to be identified based on the target bounding box of the mouse hole in multiple historical drone images.

[0046] The mouse hole targets in the dataset differ somewhat from those in natural scene datasets. To optimize the YOLOv8n model structure and improve the real-time performance of mouse hole detection, Python was used to perform statistical analysis on the size of the prior bounding boxes in the training set. The prior bounding boxes are the target boxes from the data annotation; analyzing these prior bounding boxes is crucial to determining the target scale and subsequently pruning redundant detection heads in the model. The results are as follows... Figure 4 As shown. By Figure 4 The prior bounding box width and height are distributed within a region of 0.1×0.1, that is, all within 30×30 pixels. The targets in the dataset are all small targets.

[0047] S3: Based on the target scale, determine the target detection head and neck feature integration module in the head network of the original YOLOv8n model.

[0048] YOLOv8n is the smallest model in the YOLOv8 series, making it easy to integrate into mobile devices such as drones for mousehole detection tasks. A simplified diagram of the original YOLOv8n structure is shown below. Figure 3 As shown.

[0049] The ConvMoudle, C2f, and SPPF modules in the backbone network extract features from the input image. These features are then integrated through the Concat and Upsample operations in the neck network, and finally, object detection is performed by three detector heads in the head network. The pixel sizes of these three detector heads are 20×20, 40×40, and 80×80, respectively. These different sizes are designed to accommodate targets of varying scales in the dataset images from natural scenes. 20×20, 40×40, and 80×80 represent the size of the input detector head feature maps. Smaller feature maps retain only the features of large targets, ignoring the features of smaller targets; therefore, smaller feature maps can only be used to detect large targets.

[0050] This invention optimizes the YOLOv8n network structure based on the size of the prior bounding boxes in the labeled data. It determines the target detection head and the target feature integration module in the neck region of the original YOLOv8n model based on the target scale. Redundant feature integration modules (Contact, Upsample, and C2F) responsible for large and medium targets in the head and corresponding neck regions are removed. Simultaneously, the deep feature information from the third C2F layer and the SPPF layer in the backbone network is retained and fused into the feature information of the second C2F layer through the Upsample and Concat modules. This ensures that this layer contains both the shallow geometric information needed for small target detection and the deep semantic information, efficiently and fully utilizing both shallow and deep information, thus improving the YOLOv8n model structure. Figure 5 As shown.

[0051] Reference Figure 6 The original and improved YOLOv8n models were used to identify mouse holes in a desert steppe. The visualization results showed that the original YOLOv8n could not identify the mouse holes marked with red rectangles, resulting in missed detections; the improved YOLOv8n could accurately identify all the mouse holes in this sample.

[0052] After removing detectors for large and medium-sized targets, the number of layers in Yolov8n was significantly reduced from 168 to 116, resulting in a 50% decrease in the number of parameters. The model size also decreased from 5.9 MB to 3.2 MB. These parameter reductions effectively alleviated the computational burden, increasing the detection speed from 138.9 FPS to 172.4 FPS. The reduction in model complexity did not lead to a decrease in detection performance; the P-value and AP values ​​increased to 94.6% and 98.6%, respectively. This demonstrates that removing detectors for large and medium-sized targets and fusing deep features into shallow layers reduced network complexity while simultaneously improving accuracy and average precision, proving the rationality and effectiveness of the optimization.

[0053] This invention uses Python to statistically analyze the size of prior boxes in the mouse hole dataset. Based on the statistical results, it optimizes the detection head of Yolov8n, removing redundant detection heads and most feature integration modules in the corresponding neck network. The deep feature information of the backbone network is then fused into the remaining detection heads, reducing network complexity while improving accuracy.

[0054] In summary, this invention, taking into account the correlation between UAV flight altitude and target size, proposes a high-precision and real-time grassland rodent burrow identification method: statistically annotating the size of the prior bounding boxes; optimizing the detection head of YOLOv8n based on the statistical results, eliminating redundant detection heads and most feature integration modules in the corresponding neck network; and fusing the deep feature information of the backbone network into the remaining detection heads, ultimately achieving efficient and high-precision identification of grassland rodent burrows.

[0055] Based on the same concept, the present invention also provides a grassland rodent burrow identification device, including an acquisition module, a determination module, a construction module and an identification module.

[0056] The acquisition module is used to acquire multiple historical drone images of the grassland area.

[0057] The determination module is used to determine the target scale of the mouse hole to be identified based on the target bounding box of the mouse hole in multiple historical UAV images. Based on the target scale, the target detection head and neck feature integration module is determined in the head network of the original YOLOv8n model.

[0058] The building module is used to remove the remaining detection heads other than the target detection heads in the head network of the original YOLOv8n model, remove the remaining feature integration modules other than the target feature integration module in the neck network, and retain the structure in the backbone network of the original YOLOv8n model where the third C2f layer and SPPF layer are connected to the second C2f layer in the neck network through the Upsample module and the Concat module, thus obtaining the improved YOLOv8n model.

[0059] The recognition module is used to collect images of drones to be identified in the grassland area. After cropping the images, the drone images are input into the improved YOLOv8n model to obtain the corresponding mouse hole recognition results.

[0060] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described grassland rodent burrow identification method.

[0061] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described grassland rodent burrow identification method.

[0062] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0063] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying grassland rodent burrows, characterized in that, Includes the following steps: Acquire multiple historical drone images of the grassland area; The target scale of the mouse hole to be identified is determined by the target bounding box of the mouse hole in multiple historical UAV images. Based on the target scale, the target detection head and the target feature integration module in the neck of the original YOLOv8n model are determined. The remaining detection heads other than the target detection head in the head network of the original YOLOv8n model are removed, and the remaining feature integration modules other than the target feature integration module in the neck network are removed. The structure in which the third C2f layer and SPPF layer in the backbone network of the original YOLOv8n model are connected to the second C2f layer in the neck network through the Upsample module and the Concat module is retained to obtain the improved YOLOv8n model. Images of drones to be identified are collected in the grassland area. After cropping the images, they are input into the improved YOLOv8n model to obtain the corresponding mouse hole identification results. The process of determining the target scale of the mouse hole to be identified based on the target bounding box of the mouse hole in multiple historical drone images includes the following steps: Each historical drone image is cropped to obtain multiple cropped images. Mouse holes in the cropped images are labeled to obtain multiple bounding boxes. The size of multiple target bounding boxes is statistically analyzed to obtain the target scale of the mouse hole to be identified.

2. The grassland rodent burrow identification method as described in claim 1, characterized in that, The historical drone image has a resolution of 5280 pixels × 3956 pixels, the multiple cropped images have a resolution of 300 pixels × 300 pixels, the width and height of the multiple target boxes are distributed within an area of ​​0.1 × 0.1, and the target size of the mouse hole to be identified is within 30 × 30 pixels.

3. The grassland rodent burrow identification method as described in claim 1, characterized in that, The remaining detectors, excluding the target detectors in the head network of the original YOLOv8n model, are removed. These remaining detectors include 20×20 and 40×40 detectors.

4. The grassland rodent burrow identification method as described in claim 2, characterized in that, Before cropping the drone image to be identified and inputting it into the improved YOLOv8n model, the improved YOLOv8n model needs to be trained. The training process includes the following steps: Multiple historical drone images are filtered, and the filtered historical drone images are cropped to obtain multiple sample images, each of which has mouse holes. The mouse holes in multiple sample images are labeled to obtain the dataset, which is then divided into training, testing and validation sets in an 8:1:1 ratio. The improved YOLOv8n model is trained using the training set, and tested and validated using the test and validation sets.

5. A grassland rodent burrow identification device, characterized in that, include: The acquisition module is used to acquire multiple historical drone images of the grassland area; The determination module is used to determine the target scale of the mouse hole to be identified based on the target bounding box of the mouse hole in multiple historical UAV images, and to determine the target detection head and neck feature integration module in the head network of the original YOLOv8n model based on the target scale. The building module is used to remove the remaining detection heads other than the target detection heads in the head network of the original YOLOv8n model, remove the remaining feature integration modules other than the target feature integration module in the neck network, and retain the structure in the backbone network of the original YOLOv8n model where the third C2f layer and SPPF layer are connected to the second C2f layer in the neck network through the Upsample module and the Concat module, thus obtaining the improved YOLOv8n model; The recognition module is used to collect images of drones to be identified in the grassland area, crop the images of the drones to be identified and input them into the improved YOLOv8n model to obtain the corresponding mouse hole recognition results; The process of determining the target scale of the mouse hole to be identified based on the target bounding box of the mouse hole in multiple historical drone images includes the following steps: Each historical drone image is cropped to obtain multiple cropped images. Mouse holes in the cropped images are labeled to obtain multiple bounding boxes. The size of multiple target bounding boxes is statistically analyzed to obtain the target scale of the mouse hole to be identified.

6. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the grassland rodent burrow identification method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the grassland rodent burrow identification method according to any one of claims 1-4.

Images