CN-121982587-A - Unmanned aerial vehicle-based forest vegetation detection method and system

CN121982587ACN 121982587 ACN121982587 ACN 121982587ACN-121982587-A

Abstract

The application relates to the technical field of image detection, in particular to a forest vegetation detection method and system based on an unmanned aerial vehicle, wherein the method comprises the steps of acquiring RGB images of a forest area in real time through the unmanned aerial vehicle and preprocessing the RGB images; the method comprises the steps of constructing comprehensive feature indexes of image blocks in an image, adaptively setting a contrast limiting threshold value of each image block according to the comprehensive feature indexes, locally enhancing by adopting a CLAHE algorithm, training a decay dead wood detection model of an enhanced RGB image, identifying the decay dead wood in the RGB image acquired by an unmanned aerial vehicle, and deploying the trained model to the unmanned aerial vehicle for real-time detection and position feedback. The application aims at the distinguishing characteristics of the decayed dead wood and the background surface soil in different blocks in the image so as to effectively strengthen the image and improve the accuracy of detecting and identifying the subsequent forest vegetation.

Inventors

SUN HAO
JIANG MINGYUAN
ZHANG SHENGYUAN
SUN GUANGXIN
XU TIANXIANG

Assignees

青岛科技大学
青岛海科虚拟现实研究院

Dates

Publication Date: 20260505
Application Date: 20260123

Claims (10)

1. The forest vegetation detection method based on the unmanned aerial vehicle is characterized by comprising the following steps of: Firstly, acquiring RGB images of a forest area in real time through an unmanned aerial vehicle carrying a camera and preprocessing the RGB images; dividing an RGB image into a plurality of image blocks, and constructing a comprehensive characteristic index based on the characteristic value of the LBP equivalent set of pixel points in the image blocks and the brightness difference between adjacent rows and adjacent columns in the image blocks, wherein the comprehensive characteristic index is used for representing the comprehensive characteristic expression condition of the image blocks under different illumination influences; step three, self-adaptively setting a contrast limiting threshold value of each image block according to the comprehensive characteristic index, carrying out local enhancement by adopting a CLAHE algorithm, and outputting an enhanced RGB image; Training the reinforced RGB image with a decay dead wood detection model for identifying the decay dead wood in the RGB image acquired by the unmanned aerial vehicle, and deploying the trained decay dead wood detection model to the unmanned aerial vehicle for real-time detection and position feedback.
2. The unmanned aerial vehicle-based forest vegetation detection method according to claim 1, wherein the comprehensive characteristic indexes respectively have positive correlation with characteristic values of all LBP equivalent sets meeting a preset number of conditions in the image block, brightness differences among all adjacent rows and all adjacent columns in the image block, and brightness difference confusion degrees thereof.
3. The unmanned aerial vehicle-based forest vegetation detection method of claim 2, wherein the feature values of the LBP equivalent set satisfying the preset number of conditions are determined by calculating the mean value of the distances between the neighborhood LBP feature vectors of all the pixel points in the LBP equivalent set and the standard deviation of the distances.
4. The unmanned aerial vehicle-based forest vegetation detection method of claim 3, wherein the neighborhood LBP feature vector of the pixel is obtained by sequentially arranging LBP values of all the pixels in the neighborhood of the pixel in a sequence from left to right and from top to bottom.
5. The unmanned aerial vehicle-based forest vegetation detection method of claim 4, wherein the pixel neighborhood is a3 x 3 window centered on a pixel.
6. The unmanned aerial vehicle-based forest vegetation detection method of claim 2, wherein the difference in luminance between adjacent rows is determined by the distance of the luminance values between adjacent rows and the difference in luminance between adjacent columns is determined by the distance of the luminance values between adjacent columns.
7. The unmanned aerial vehicle-based forest vegetation detection method of claim 2, wherein the segmentation threshold value of the brightness value of all pixels in the RGB image is obtained before the comprehensive characteristic index is calculated; Taking the ratio of the number of pixels in the image block, which is larger than or equal to the brightness value of the segmentation threshold, to the number of pixels in the image block as the weight of the characteristic values of all LBP equivalent sets meeting the preset number conditions in the image block when the comprehensive characteristic index is constructed; and taking the ratio of the number of pixels in the image block, which is smaller than the brightness value of the segmentation threshold, to the number of pixels in the image block as the weight of brightness differences and the brightness difference confusion degree of all adjacent lines in the image block when the comprehensive characteristic index is constructed.
8. The unmanned aerial vehicle-based forest vegetation detection method of claim 1, wherein the contrast limit threshold is determined by linearly mapping the normalized composite characteristic index to [ [ After the interval is determined, 、 Limiting threshold value for preset maximum and minimum contrast, and 。
9. The unmanned aerial vehicle-based forest vegetation detection method of claim 1, wherein the data set trained by the decay dead wood detection model is a data set consisting of a plurality of images collected by unmanned aerial vehicle historic through label marking, and the marked labels are two types of decay dead wood and healthy plants.
10. An unmanned aerial vehicle-based forest vegetation detection system comprising a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the unmanned aerial vehicle-based forest vegetation detection method of any of claims 1-9.

Description

Unmanned aerial vehicle-based forest vegetation detection method and system Technical Field The application relates to the technical field of image detection, in particular to a forest vegetation detection method and system based on an unmanned aerial vehicle. Background The traditional detection and management of forest vegetation generally has low detection efficiency due to large workload, and along with the development of technology, unmanned aerial vehicle-based forest vegetation detection gradually becomes an indispensable efficient mode in current forest vegetation detection and management. In the forest vegetation detection operation based on unmanned aerial vehicle, consider that the forest can produce the wood turnover (including dead wood) under the condition of being interfered by a plurality of factors, wherein the interference factors comprise the natural death, the environment and biological interference (such as wind disaster and insect diseases) of the forest, the wood is decayed and embrittled along with the influence of the insect diseases, the wind resistance capability is greatly attenuated, and the situation that the decay dead wood possibly breaks and turnover under the influence of strong wind can be formed. In the process of identifying the rotten and dead wood, bark is fallen off due to the influence of diseases and insect pests, the colors are changed from light gray, yellow brown to dark brown and even black, under the condition of poor light, the characteristics of the rotten and dead wood and the characteristics of the ground surface soil are similar, the conventional image enhancement algorithm generally enhances the overall pixel characteristics, the characteristic distinction between the rotten and dead wood and the ground surface soil cannot be effectively enhanced, and the situation of misjudgment and missed detection easily occurs when forest vegetation is detected and identified. Disclosure of Invention In order to solve the technical problems, the application provides a forest vegetation detection method and system based on an unmanned aerial vehicle, and the adopted technical scheme is as follows: in a first aspect, an embodiment of the present application provides an unmanned aerial vehicle-based forest vegetation detection method, including the steps of: Firstly, acquiring RGB images of a forest area in real time through an unmanned aerial vehicle carrying a camera and preprocessing the RGB images; dividing an RGB image into a plurality of image blocks, and constructing a comprehensive characteristic index based on the characteristic value of the LBP equivalent set of pixel points in the image blocks and the brightness difference between adjacent rows and adjacent columns in the image blocks, wherein the comprehensive characteristic index is used for representing the comprehensive characteristic expression condition of the image blocks under different illumination influences; step three, self-adaptively setting a contrast limiting threshold value of each image block according to the comprehensive characteristic index, carrying out local enhancement by adopting a CLAHE algorithm, and outputting an enhanced RGB image; Training the reinforced RGB image with a decay dead wood detection model for identifying the decay dead wood in the RGB image acquired by the unmanned aerial vehicle, and deploying the trained decay dead wood detection model to the unmanned aerial vehicle for real-time detection and position feedback. Preferably, the comprehensive feature indexes respectively have positive correlation with feature values of all LBP equivalent sets meeting the preset quantity conditions in the image block, brightness differences among all adjacent rows and all adjacent columns in the image block and brightness difference confusion degrees thereof. Preferably, the feature values of the LBP equivalent set meeting the preset number of conditions are determined by calculating the average value of the distances and the standard deviation of the distances between the neighbor LBP feature vectors of all the pixel points in the LBP equivalent set. Preferably, the neighborhood LBP feature vector of the pixel point is obtained by sequentially arranging LBP values of all pixel points in the neighborhood of the pixel point in sequence from left to right and from top to bottom. Preferably, the pixel neighborhood is a3×3 window centered on the pixel. Preferably, the luminance difference between adjacent rows is determined by the distance of the luminance values between adjacent rows and the luminance difference between adjacent columns is determined by the distance of the luminance values between adjacent columns. Preferably, before calculating the comprehensive characteristic index, the segmentation threshold value of the brightness value of all pixel points in the RGB image is obtained; Taking the ratio of the number of pixels in the image block, which is larger than or equal to the b