CN-122023966-A - Method and system for constructing alpine grassland rat hole and bare spot recognition model based on UAV (unmanned aerial vehicle) image and deep learning

Abstract

The invention belongs to the technical field of unmanned aerial vehicle low-altitude remote sensing, and particularly relates to a method and a system for constructing a recognition model of a rat hole and a bare spot of a alpine grassland based on UAV (unmanned aerial vehicle) images and deep learning. Through optimizing unmanned aerial vehicle flight collection strategy and combining with lightweight deep learning model, the full-flow low-power consumption and quick response technical scheme from data acquisition to analysis result is constructed, hardware cost and field investigation period are reduced, and the defects of high cost, long period and poor precision of traditional manual investigation and monitoring are overcome. A private data set is built aiming at a specific monitoring scene of the alpine grassland, a highly light-weight special scene recognition model is trained, the limitations of poor effect and high resource consumption of a general remote sensing model in the specific scene can be overcome, and high-precision rapid recognition and detection can be still kept on a low-cost computing platform. By simplifying the data processing flow, the technical operation threshold and the implementation complexity are reduced, and a feasible solution is provided for monitoring the grassland mouse damage in a small range at high frequency.

Inventors

• Sun feida
• HAN BINRU
• ZHANG HUANHUAN
• BAI YANFU
• LIU LIN
• ZHOU JIQIONG
• MA ZHOUWEN
• YANG TINGYONG
• YAO JIANMIN

Assignees

• 四川农业大学

Dates

Publication Date

20260512

Application Date

20260203

Claims (7)

1. 1. A method for constructing a recognition model of a rat hole and a bare spot in alpine grassland based on UAV image and deep learning is characterized by comprising the following steps: S1, unmanned aerial vehicle image acquisition, namely respectively carrying out low-altitude RGB image acquisition on a target alpine grassland in a warm season of 8 months and a cold season of 11 months by using an unmanned aerial vehicle; s2, constructing an image segmentation dataset, namely extracting sample data about a rat hole and a bare spot in an acquired RGB image, marking and generating class polygon boundary coordinates, class pixel labels and masks of the rat hole and the bare spot through Labelme, and constructing a Pascal VOC (volatile organic compound) format image segmentation dataset; S3, model construction, namely reasoning the image segmentation dataset by taking MobileNet as a main network to obtain a alpine grassland rat hole and bare spot segmentation model based on UAV images and deep learning.
2. 2. The method for constructing a model according to claim 1, wherein the unmanned aerial vehicle aerial photographing height is less than or equal to 20m during the image acquisition in S1, and the photographing angle is 180 ° in top view.
3. 3. The method according to claim 1, wherein S2 the image segmentation dataset construction is specifically: image cutting, which is to cut the RGB image with original size into 512X 512 pixel size; Marking a class sample polygon boundary of two types of targets, namely a rat hole and a bare spot, by adopting LabelMe to generate a JSON file containing class labels and boundary coordinates, namely a coordinate point sequence of the polygon, and taking the coordinate point sequence as a vector format; mask generation, namely converting a JSON coordinate file into a Mask containing class boundary pixel labels, namely pixel level representation of polygons, in a grid mode by adopting a Labelme2voc.py script in Labelme; The label class is that the background pixel class is specified to be 0, the bare spot pixel class is specified to be 1, and the mousehole pixel class is specified to be 2.
4. 4. A alpine grassland rat hole and nude spot recognition model based on UAV image and deep learning obtained according to the model construction method of any one of claims 1-3.
5. 5. The alpine grassland rat hole and bare spot recognition system based on UAV image and deep learning is characterized by comprising the following modules: the image acquisition module is used for acquiring target cold grassland images; the data processing module is used for preprocessing the acquired images; the segmentation labeling module is used for labeling the image segmentation categories of the preprocessed images; and the output module is used for processing and dividing rat holes and bare spot areas of the target image for the extracted features.
6. 6. The alpine grassland rat hole and bare spot recognition method based on UAV image and deep learning is characterized in that the method inputs an unmanned aerial vehicle image to recognize the rat hole and the bare spot through the model of claim 4 or the system of claim 5.
7. 7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor for executing the model of claim 4 or the system of claim 5.

Description

Method and system for constructing alpine grassland rat hole and bare spot recognition model based on UAV (unmanned aerial vehicle) image and deep learning Technical Field The invention belongs to the technical field of unmanned aerial vehicle low-altitude remote sensing, and particularly relates to a method and a system for constructing a recognition model of a rat hole and a bare spot of a alpine grassland based on UAV (unmanned aerial vehicle) images and deep learning. Background The classification of features of the grassland ecosystem and its dynamic monitoring are important bases for evaluating grassland degradation. The ground object categories can be divided into four categories of elements of vegetation, nude spots, mouseholes and other categories according to the ground cover characteristics. The rat hole distribution density and the bare spot area (bald spot rate) are used as key quantization indexes, and are established as core basis for the division of a rat damage control threshold value and a rat barren land grading standard in the industry standard of the agricultural rural department, the grassland and barren land treatment technical specification and the standard of the Sichuan province, the Qinghai-Tibet plateau grassland ground rat control technical specification. Along with the iteration upgrading of the remote sensing technology, the unmanned aerial vehicle low-altitude remote sensing has formed a technical system covering 'animal and plant collaborative monitoring' by virtue of the high space-time resolution advantage, wherein in animal monitoring dimension, the unmanned aerial vehicle low-altitude remote sensing focuses on the dynamic tracking of rodent population except for the conventional wild animal protection research, a mouse damage early warning system is constructed by technical means such as hole group identification, suitable living area modeling and the like, and in a grassland ground monitoring layer, the accurate measurement and calculation of the baldness rate are realized by multispectral image fusion analysis. At present, relevant researches on grassland mouse damage monitoring based on unmanned aerial vehicle remote sensing mainly pay attention to aspects of mouse hole recognition based on algorithms such as target detection, ground surface mouse damage information extraction fused with multi-source remote sensing data, rodent suitable living area research combining geographic information with ecology and the like, and the exploration of the technologies and methods provides a new methodology support for grassland degradation mechanism analysis and ecological restoration. However, at present, no model for effectively identifying rat holes and bare spots in alpine grasslands exists, and the manual identification and monitoring have low efficiency and large limitation on the scale. Disclosure of Invention Based on the problems, the application provides a method and a system for constructing a recognition model of a rat hole and a bare spot in alpine grassland based on UAV (unmanned aerial vehicle) images and deep learning. By constructing a rat hole target detection and bare spot semantic segmentation model, the distribution states and relative area occupation ratios of the grass rat holes and the bare spots are inferred by adopting RGB images, so that the bottleneck of traditional manual investigation is broken through, the development of a grass rat damage accurate monitoring and early warning technology system is assisted, and technical support is provided for dynamic monitoring, accurate early warning and prevention and control decision. The first technical scheme of the application discloses a method for constructing a recognition model of a rat hole and a bare spot of a alpine grassland based on UAV (unmanned aerial vehicle) images and deep learning, which comprises the following steps: S1, unmanned aerial vehicle image acquisition, namely respectively carrying out low-altitude RGB image acquisition on a target alpine grassland in a warm season of 8 months and a cold season of 11 months by using an unmanned aerial vehicle; s2, constructing an image segmentation dataset, namely extracting sample data about a rat hole and a bare spot in an acquired RGB image, marking and generating class polygon boundary coordinates, class pixel labels and masks of the rat hole and the bare spot through Labelme, and constructing a Pascal VOC (volatile organic compound) format image segmentation dataset; S3, model construction, namely reasoning the image segmentation dataset by taking MobileNet as a main network to obtain a alpine grassland rat hole and bare spot segmentation model based on UAV images and deep learning. Furthermore, in the step S1, the aerial photographing height of the unmanned aerial vehicle is less than or equal to 20m, and the photographing angle is 180 degrees in overlook. Further, the image segmentation dataset construction in S2 specifically includes: image c