CN-122024068-A - Vegetation detection method, system, computer equipment and storage medium

CN122024068ACN 122024068 ACN122024068 ACN 122024068ACN-122024068-A

Abstract

The invention provides a vegetation detection method, a system, computer equipment and a storage medium, which belong to the field of remote sensing data processing, and comprise the steps of collecting vegetation UAV image data according to 5 different voyages in a sample area, collecting the vegetation UAV image data according to a single voyage in a research area, marking positions and quantity of vegetation targets on the vegetation UAV image data collected in the sample area, sequentially carrying out Gaussian fuzzy simulation and area resampling on the marked image data to obtain a plurality of groups of simulated image data consistent with the image resolution characteristics of the research area, respectively training target detection models through the plurality of groups of simulated image data, inputting the vegetation UAV image data of the actual single voyage in the collected research area into the trained 5 groups of target detection models, outputting respective vegetation detection frame coordinates and confidence results, fusing the vegetation detection frames and confidence results, and outputting vegetation numbers corresponding to the single-altitude images in the research area. The method realizes the cooperative unification of large-scale coverage and high-precision identification.

Inventors

DUAN DONGWEI
Yao Dongze
HE XIAOLANG
SUN PINGPING
DING XIANG
CHEN SHI
PU ZHIGUO
DU ZHEN

Assignees

中煤能源研究院有限责任公司
西安交通大学

Dates

Publication Date: 20260512
Application Date: 20260224

Claims (10)

1. The vegetation detection method is characterized by comprising the following steps of: determining a sample area and a research area of target vegetation, wherein the area of the sample area is smaller than that of the research area; Collecting vegetation UAV image data according to at least 5 different voyages in the sample area, and collecting vegetation UAV image data according to a single voyage in the research area, wherein different sampling voyages in the sample area are smaller than those in the research area; Labeling positions and quantity of vegetation targets on the vegetation UAV image data acquired by the sample area, and sequentially performing Gaussian fuzzy simulation and area resampling processing on the labeled image data to obtain a plurality of groups of simulated image data with the same navigation height of the sample areas and single navigation height image resolution characteristics of the research areas; Training the target detection models through multiple groups of simulation image data respectively to obtain at least 5 sets of trained target detection models; The method comprises the steps of inputting acquired vegetation UAV image data of a real single aerial height of a research area into 5 sets of trained target detection models, outputting respective vegetation detection frames and confidence results by each set of models, carrying out fusion processing on the vegetation detection frames and the confidence results output by each set of models, and outputting vegetation plants corresponding to the single aerial height image of the research area.
2. A vegetation detection method according to claim 1, wherein the sample areas have different elevations of 10m, 20m, 50m, 100m and 200m, respectively, and the single elevation of the investigation region is 500m.
3. The vegetation detection method of claim 2 wherein the vegetation UAV image data comprises near infrared NIR, red edge RedEdge, red band, green band, NDVI, and RGB images.
4. A vegetation detection method according to claim 3, wherein the steps of sequentially performing gaussian blur simulation and area resampling on the labeled image data to obtain a plurality of groups of simulated image data with the same navigation height of the plurality of sample areas and the single navigation height image resolution characteristic of the research area, specifically include the following steps: calculating a fuzzy coefficient by using a formula sigma=kappa.500/h through Gaussian fuzzy simulation of detail loss of high-altitude imaging, wherein h is a low-altitude, and kappa is an adjusting parameter; And resampling the blurred image according to the ratio of r=500/h through reducing the resolution of the area resampling analog high-altitude imaging, merging the low-navigation high-resolution pixels into the low-resolution pixels with the navigation height of 500m, completely matching the resolution scale of the real 500m image, and obtaining five groups of analog images, wherein each group of analog images contains G, R, red-edge, NIR, NDVI, RGB channel data.
5. The vegetation detection method according to claim 1, wherein a weighted frame fusion algorithm is adopted to fuse the vegetation detection frames and confidence results output by each set of models, and the vegetation plants corresponding to the single aerial image in the research area are output, and specifically comprising the following steps: Traversing detection frames output by all models, and calculating the cross-over ratio between the frames; when the intersection ratio IoU of two or more detection frames is more than 0.5, determining the same target and classifying the same target into a cluster; carrying out weighted average on a plurality of detection frames which are judged to be the same target according to the confidence coefficient to obtain a final detection frame, namely, the confidence coefficient is weighted and fused with WBF; the detection frames which do not meet the overlapping condition are reserved as independent targets, and meanwhile, the low confidence coefficient result is removed according to a preset threshold value; After the repeated screening, outputting vegetation final detection frame coordinates and plant numbers corresponding to 500m aerial height images of the research area, and synchronously outputting fusion confidence.
6. A vegetation detection method according to claim 1, wherein the target detection model is YOLOv model or YOLOv5 model.
7. The vegetation detection method of claim 1, wherein after the vegetation UAV image data is acquired at each stage, the vegetation UAV image data is first subjected to a radiation correction and geometric registration pre-process.
8. A vegetation detection system, comprising: The data acquisition module is used for determining a sample area and a research area of target vegetation, and the area of the sample area is smaller than that of the research area; Collecting vegetation UAV image data according to at least 5 different voyages in the sample area, and collecting vegetation UAV image data according to a single voyage in the research area, wherein different sampling voyages in the sample area are smaller than those in the research area; The data processing module is used for marking the positions and the number of vegetation targets on the vegetation UAV image data acquired by the sample area, and sequentially carrying out Gaussian fuzzy simulation and area resampling processing on the marked image data to obtain a plurality of groups of simulated image data with the same navigational height of the sample area as that of the single navigational high image resolution characteristic of the research area; The model training module is used for training the target detection models through a plurality of groups of simulation image data respectively to obtain at least 5 sets of trained target detection models; The vegetation detection module is used for inputting acquired real single-voyage vegetation UAV image data of the research area into 5 sets of trained target detection models, outputting respective vegetation detection frames and confidence results by each set of models, carrying out fusion processing on the vegetation detection frames and the confidence results output by each set of models, and outputting vegetation plants corresponding to the single-voyage images of the research area.
9. A computer device comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to carry out the steps of the method according to any one of claims 1 to 7.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when loaded by a processor, is able to carry out the steps of the method according to any one of claims 1 to 7.

Description

Vegetation detection method, system, computer equipment and storage medium Technical Field The invention belongs to the technical field of remote sensing data processing, and particularly relates to a vegetation detection method, a vegetation detection system, computer equipment and a storage medium. Background In the field of monitoring of specific species (bushes, trees and the like) in large-scale areas (such as hundreds of square kilometers), accurate grasp of information such as species distribution, coverage and the like has important significance for ecological protection, resource management and engineering environment assessment (such as ecological restoration monitoring in coal mine areas). The area of the area is wide, and the area is limited by natural conditions such as weather, topography and the like, and the monitoring work needs to simultaneously consider coverage and recognition accuracy so as to provide reliable data support for subsequent decisions. However, under the current technical conditions, the cooperative implementation of large scale and high precision still faces significant challenges, and the traditional monitoring mode has difficulty in balancing the core requirements of efficiency and accuracy. The main current monitoring methods are mainly divided into two types, namely low-altitude high-resolution unmanned aerial vehicle aerial survey (flight height is below 30 m), and the main current monitoring methods have the advantages of being capable of clearly distinguishing single plants or small plaque species, extremely high in recognition precision and capable of meeting the fine monitoring requirements, but have the disadvantages of small single course coverage area, strict requirements on solar altitude angle in multispectral flight, incapability of operation in rainy days, strong winds and other weather conditions, extremely long monitoring operation time in large-area areas, high cost and difficulty in realizing frequent monitoring and large-scale popularization. The other type is aerial survey at high altitude or satellite remote sensing data (such as aerial survey of unmanned aerial vehicle with flight height of 300-500m, medium-low resolution satellite data), the method has the core advantages of large coverage area and high operation efficiency, can rapidly complete monitoring coverage of a large-scale area, is limited by resolution, has low recognition precision on target species, and cannot meet the requirements of fine recognition and coverage estimation. The existing monitoring work always faces the two difficult trade-offs of high precision, small range and low precision and large range, and an effective technical path capable of taking advantages of the two technologies into consideration is lacking. To make up for the gap, some researches try to improve the recognition accuracy through algorithm optimization, for example, on the basis of single-height or single-resolution images, image quality and classification effect are improved by adopting means such as image enhancement, super-resolution reconstruction or traditional classification algorithm. However, the essence of the method is still based on optimization of single resolution data, the inherent resolution correlation between the high-altitude high-resolution sample and the high-altitude low-resolution image cannot be fully utilized, and the problem of detail information deletion of the target species in the high-altitude low-resolution image cannot be fundamentally solved, so that the improvement effect on the coverage estimation of the target species is very limited, and the actual requirement of high-precision monitoring of a large-scale area is difficult to meet. Disclosure of Invention The invention provides a vegetation detection method, which aims to solve the problem that the existing monitoring method can not fundamentally solve the problem that the detail information of a target species in a high-altitude low-resolution image is missing. In order to achieve the above object, the present invention provides the following technical solutions: a vegetation detection method comprising: determining a sample area and a research area of target vegetation, wherein the area of the sample area is smaller than that of the research area; Collecting vegetation UAV image data according to at least 5 different voyages in the sample area, and collecting vegetation UAV image data according to a single voyage in the research area, wherein different sampling voyages in the sample area are smaller than those in the research area; Labeling positions and quantity of vegetation targets on the vegetation UAV image data acquired by the sample area, and sequentially performing Gaussian fuzzy simulation and area resampling processing on the labeled image data to obtain a plurality of groups of simulated image data with the same navigation height of the sample areas and single navigation height image resolution characteristi