CN-121982524-A - Vegetation coverage rapid measurement method and system based on image recognition

CN121982524ACN 121982524 ACN121982524 ACN 121982524ACN-121982524-A

Abstract

The invention relates to the technical field of ecological intelligent monitoring, in particular to a vegetation coverage rapid measurement method and a vegetation coverage rapid measurement system based on image identification, wherein the method comprises the following steps of constructing a multi-source vegetation image acquisition device, and acquiring vegetation image initial data of a target area based on the multi-source vegetation image acquisition device; the method comprises the steps of carrying out image preprocessing and illumination self-adaptive correction on initial vegetation image data to obtain standardized vegetation image data, establishing an improved lightweight semantic segmentation model, segmenting the standardized vegetation image data to obtain vegetation image segmentation results, carrying out preliminary calculation on vegetation coverage of the target area based on the vegetation image segmentation results to obtain preliminary vegetation coverage, constructing a geometric distortion dynamic compensation model to dynamically compensate the preliminary vegetation coverage, and realizing rapid calculation on the vegetation coverage. The invention effectively improves the efficiency and accuracy of vegetation coverage measurement.

Inventors

YANG JINJIE
JI HONGCHAO
YU YANG
LIU HAIJIANG
DONG GUIHUA

Assignees

中国环境监测总站

Dates

Publication Date: 20260505
Application Date: 20260115

Claims (10)

1. The vegetation coverage rapid measurement method based on image recognition is characterized by comprising the following steps: Constructing a multi-source vegetation image acquisition device, and acquiring initial vegetation image data of a target area based on the multi-source vegetation image acquisition device; Performing image preprocessing and illumination self-adaptive correction on the vegetation image initial data to obtain vegetation image standardized data; establishing an improved lightweight semantic segmentation model, and segmenting the vegetation image standardized data to obtain a vegetation image segmentation result; performing preliminary measurement and calculation on the vegetation coverage of the target area based on the vegetation image segmentation result to obtain preliminary vegetation coverage; and constructing a geometric distortion dynamic compensation model to dynamically compensate the preliminary vegetation coverage so as to realize quick measurement and calculation of the vegetation coverage.
2. The rapid vegetation coverage measuring method based on image recognition according to claim 1, wherein the constructing a multi-source vegetation image acquisition device, acquiring initial vegetation image data of a target area based on the multi-source vegetation image acquisition device, comprises: constructing the multi-source vegetation image acquisition device by combining an unmanned plane and ground shooting equipment; obtaining long-range vegetation image data of the target area according to the unmanned aerial vehicle; collecting close-range vegetation image data of the target area according to the ground shooting equipment; And taking the long-range vegetation image data and the short-range vegetation image data as the vegetation image initial data.
3. The method for rapidly measuring vegetation coverage based on image recognition according to claim 1, wherein the performing image preprocessing and illumination adaptive correction on the initial vegetation image data to obtain the standardized vegetation image data comprises: Performing image preprocessing on the vegetation image initial data to obtain vegetation image preprocessing data, wherein the image preprocessing comprises image format and resolution standardization, image denoising, edge enhancement and image registration; Filtering a non-vegetation region of the vegetation image preprocessing data to obtain vegetation image data of the target region; And obtaining illumination non-uniformity quantization factors to construct an illumination dynamic correction function, and combining the vegetation image data and the illumination dynamic correction function to obtain the vegetation image standardized data.
4. The rapid vegetation coverage measurement method based on image recognition according to claim 1, wherein the building of the improved lightweight semantic segmentation model comprises: Constructing a dynamic fusion gating mechanism, and building a double-branch heterogeneous trunk model structure by combining a space branch and a spectrum branch; Obtaining a vegetation image density priori index based on the vegetation image standardized data, and constructing a pruning rate control function according to the vegetation image density priori index; And establishing a deformable convolution attention unit, and constructing the improved lightweight semantic segmentation model by combining the dual-branch heterogeneous trunk model structure and the pruning rate control function.
5. The rapid vegetation coverage measuring method based on image recognition according to claim 4, wherein the segmenting the vegetation image standardization data to obtain vegetation image segmentation results comprises: acquiring vegetation space morphological characteristics according to the space branch based on the vegetation image standardized data, and acquiring vegetation spectral response characteristics according to the spectrum branch; carrying out dynamic weighted fusion on the vegetation space morphological characteristics and the vegetation spectral response characteristics according to the dynamic fusion gating mechanism to obtain a vegetation enhancement joint characteristic diagram; Performing self-adaptive feature pruning on the vegetation enhancement joint feature map according to the pruning rate control function to obtain a vegetation pruning feature map; Performing feature sensing on the vegetation pruning feature map by combining the deformable convolution attention unit to obtain a refined vegetation morphological feature map; and carrying out binarization segmentation on the refined vegetation morphological feature map to obtain a vegetation image segmentation result.
6. The rapid vegetation coverage measuring method based on image recognition according to claim 1, wherein the performing preliminary measurement on the vegetation coverage of the target area based on the vegetation image segmentation result to obtain a preliminary vegetation coverage includes: establishing a three-dimensional overlap correction mechanism, and obtaining a leaf overlap degree quantized value based on the vegetation image segmentation result; constructing a withering condition criterion, and obtaining a withering quantized value of the vegetation image segmentation result according to the withering condition criterion; and combining the vegetation image segmentation result, the leaf overlapping degree quantized value and the withered object quantized value, and carrying out differential weight distribution based on vegetation types to obtain the preliminary vegetation coverage.
7. The rapid vegetation coverage determination method based on image recognition of claim 6, wherein the constructing a litter condition criterion comprises: collecting a living vegetation sample under standard illumination conditions, and obtaining living vegetation texture characteristics and a near infrared band reflectivity baseline interval of the living vegetation sample; Correcting the near-infrared band reflectivity baseline interval according to the living vegetation texture characteristics to obtain a near-infrared band reflectivity threshold; and acquiring the near infrared band average reflectivity of the vegetation image segmentation result, and constructing the withering condition criterion by combining the near infrared band reflectivity threshold.
8. The rapid vegetation coverage measuring method based on image recognition according to claim 1, wherein the constructing a geometric distortion dynamic compensation model dynamically compensates the preliminary vegetation coverage to realize rapid measurement and calculation of the vegetation coverage, comprising: acquiring a multi-parameter distortion factor of the multi-source vegetation image acquisition device, and constructing the geometric distortion dynamic compensation model based on the multi-parameter distortion factor; performing edge region self-adaptive compensation on the vegetation image segmentation result according to the geometric distortion dynamic compensation model to obtain an edge distortion rate; And carrying out dynamic compensation on the preliminary vegetation coverage by combining the edge distortion rate to obtain the vegetation coverage.
9. The rapid image recognition-based vegetation coverage measurement method of claim 8, wherein the acquiring the multi-parameter distortion factor of the multi-source vegetation image acquisition device comprises: And deploying an inertial measurement unit on the multi-source vegetation image acquisition device, and acquiring the multi-parameter distortion factors including pitch angle, yaw angle and relative altitude through the inertial measurement unit.
10. A vegetation coverage rapid measurement system based on image recognition, comprising: the initial data module is used for constructing a multi-source vegetation image acquisition device and acquiring vegetation image initial data of a target area based on the multi-source vegetation image acquisition device; The standardized data module is used for carrying out image preprocessing and illumination self-adaptive correction on the vegetation image initial data to obtain vegetation image standardized data; The segmentation result module is used for establishing an improved lightweight semantic segmentation model and segmenting the vegetation image standardized data to obtain a vegetation image segmentation result; the preliminary vegetation coverage module is used for carrying out preliminary measurement and calculation on the vegetation coverage of the target area based on the vegetation image segmentation result to obtain preliminary vegetation coverage; And the rapid calculation module is used for constructing a geometric distortion dynamic compensation model to dynamically compensate the preliminary vegetation coverage so as to realize rapid calculation of the vegetation coverage.

Description

Vegetation coverage rapid measurement method and system based on image recognition Technical Field The invention relates to the technical field of ecological intelligent monitoring, in particular to a vegetation coverage rapid measurement method and system based on image recognition. Background In the prior art, vegetation coverage measurement and calculation mostly depends on remote sensing image analysis, ground sampling investigation or a combination mode of the remote sensing image analysis and the ground sampling investigation. In the aspect of remote sensing images, a multispectral image obtained by satellites or unmanned aerial vehicles is often utilized, vegetation areas are divided by calculating indexes such as normalized vegetation indexes, and then the vegetation pixel occupation ratio is counted to obtain coverage, and the images are classified by combining a traditional machine learning model by a part of methods, so that vegetation and non-vegetation boundaries are further refined. The ground investigation is carried out by arranging a sample side, manually measuring the vegetation coverage area in the sample side and calculating the whole situation of the area, or acquiring vegetation reflectivity data by using a handheld spectrometer for auxiliary calculation. In addition, the remote sensing image and the ground sampling data are fused according to part of the technical scheme, and the coverage space distribution result is optimized through an interpolation method so as to balance the measuring and calculating range and the precision. However, the prior art has obvious limitations, including that firstly, the remote sensing image is limited by resolution, the recognition precision of low vegetation and sparse areas is insufficient and is easily interfered by cloud layers and atmospheric scattering, secondly, the classification effect of a traditional machine learning model is poor under a complex scene, the measurement precision of vegetation coverage is not high, thirdly, the ground sampling precision is high, time and effort are consumed, the rapid measurement requirement of a large area is difficult to adapt, fourthly, the geometrical distortion caused by illumination change and equipment posture when the image acquisition is not fully considered in the prior art, and the misjudgment problem of dead objects and living vegetation is caused, so that systematic deviation exists in measurement results, and thirdly, the existing segmentation model is mostly a heavy network, the calculation efficiency is low, the actual application requirement of real-time and rapid coverage is difficult to be satisfied, and the immediate application of vegetation coverage data in the fields of measurement and calculation, agricultural management and the like is restricted. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a vegetation coverage rapid measurement method and system based on image recognition. In order to achieve the above object, in a first aspect, the present invention provides a method for rapidly measuring vegetation coverage based on image recognition, the method comprising the steps of: Constructing a multi-source vegetation image acquisition device, and acquiring initial vegetation image data of a target area based on the multi-source vegetation image acquisition device; Performing image preprocessing and illumination self-adaptive correction on the vegetation image initial data to obtain vegetation image standardized data; establishing an improved lightweight semantic segmentation model, and segmenting the vegetation image standardized data to obtain a vegetation image segmentation result; performing preliminary measurement and calculation on the vegetation coverage of the target area based on the vegetation image segmentation result to obtain preliminary vegetation coverage; and constructing a geometric distortion dynamic compensation model to dynamically compensate the preliminary vegetation coverage so as to realize quick measurement and calculation of the vegetation coverage. According to the invention, the data comprehensiveness is improved through multi-source acquisition, the data quality is ensured through preprocessing and illumination correction, the segmentation precision and speed are both considered by improving a lightweight model, the characteristic optimization is combined with preliminary measurement and calculation, the distortion influence is eliminated through dynamic compensation, and the vegetation coverage is rapidly and accurately measured and calculated. Optionally, the constructing the multi-source vegetation image acquisition device, based on the initial vegetation image data of the target area acquired by the multi-source vegetation image acquisition device, includes: constructing the multi-source vegetation image acquisition device by combining an unmanned plane and ground shooting equipment; obtaining long-range vegetation