CN-120689776-B - Unmanned aerial vehicle infrared photovoltaic image-based component segmentation method and device

CN120689776BCN 120689776 BCN120689776 BCN 120689776BCN-120689776-B

Abstract

The invention relates to the technical field of computer vision and image processing, in particular to a component segmentation method, device, equipment and storage medium based on an unmanned aerial vehicle infrared photovoltaic image, which are combined with sub-pixel structure feature reconstruction and a multi-mode gradient edge attention mechanism to effectively inhibit background noise of the unmanned aerial vehicle infrared photovoltaic image, reduce the background misjudgment rate, improve the accuracy of identifying sub-pixel level structures, thereby improving the component edge segmentation precision of the unmanned aerial vehicle infrared photovoltaic image, effectively overcoming the problems that the unmanned aerial vehicle infrared photovoltaic image is difficult to accurately segment under complex illumination and shielding scenes and easy to cause false detection, and meeting the industrial detection demands.

Inventors

HUANG XIHUA
LI ZHENHAO
HOU ZHIWEI
JING WENLONG
HU HONGDA
HUANG WUMENG
YIN CHAO
SUN JIA
YANG JI
WU XIWEN

Assignees

广东省科学院广州地理研究所
南方海洋科学与工程广东省实验室(广州)
广东工业大学

Dates

Publication Date: 20260512
Application Date: 20250527

Claims (7)

1. The component segmentation method based on the unmanned aerial vehicle infrared photovoltaic image is characterized by comprising the following steps of: The method comprises the steps of obtaining an unmanned aerial vehicle infrared photovoltaic image to be segmented and a preset photovoltaic image segmentation model, wherein the photovoltaic image segmentation model comprises a feature extraction module, a feature reconstruction module, a feature strengthening module and a component segmentation module, wherein the feature extraction module comprises a coding module and a multi-scale pyramid pooling module; inputting the unmanned aerial vehicle infrared photovoltaic image to be segmented into the coding module for coding processing to obtain a characteristic coding diagram; Inputting the characteristic coding diagram into a convolution layer of the multi-scale pyramid pooling module for convolution processing to obtain a convolution characteristic diagram, and pooling the convolution characteristic diagram serving as an input characteristic diagram of a first pooling layer to obtain a characteristic pooling diagram output by the first pooling layer; repeatedly carrying out pooling treatment until the characteristic pooling diagram output by the last pooling layer is obtained, and taking the characteristic pooling diagram as a characteristic extraction diagram to obtain a plurality of scale characteristic extraction diagrams; Inputting the feature extraction graphs of the scales into the feature reconstruction module, and carrying out up-sampling layer-by-layer fusion on the feature extraction graphs of the last scale by adopting a stepped bilinear interpolation method according to the feature extraction graphs of the scales to obtain feature fusion graphs of the scales; Carrying out sub-pixel convolution on the feature fusion map of a plurality of scales to obtain a sub-pixel convolution feature map of a plurality of scales; Inputting the unmanned aerial vehicle infrared photovoltaic image to be segmented and the characteristic reconstruction image into the characteristic strengthening module, adopting a Sobel operator to calculate a multi-mode gradient map of the unmanned aerial vehicle infrared photovoltaic image to be segmented to obtain an intensity gradient map and a texture gradient map, and fusing the intensity gradient map and the texture gradient map to obtain a composite edge response map; performing attention weight calculation according to the composite edge response graph to obtain attention weight parameters; multiplying the characteristic reconstruction graph by the attention weight parameter element by element to obtain an edge characteristic enhancement graph; Inputting the edge characteristic enhancement map into the component segmentation module to carry out component segmentation, and obtaining a component segmentation map as a component segmentation result of the unmanned aerial vehicle infrared photovoltaic image to be segmented.
2. The method for segmenting the unmanned aerial vehicle infrared photovoltaic image-based component according to claim 1, wherein the step of inputting the unmanned aerial vehicle infrared photovoltaic image to be segmented into the feature extraction module to perform multi-scale feature extraction to obtain feature extraction graphs with a plurality of scales, further comprises the steps of: Calculating a characteristic image pixel mean value and a characteristic image pixel standard deviation of the characteristic extraction image of the last scale to obtain the characteristic image pixel mean value and the characteristic image pixel standard deviation; and carrying out illumination correction on the feature extraction graph of the last scale according to the feature graph pixel mean value and the feature image pixel standard deviation to obtain the feature extraction graph of the last scale after processing.
3. The unmanned aerial vehicle infrared photovoltaic image-based component segmentation method according to claim 1, further comprising the steps of: And obtaining the optimized and post-processed component segmentation diagram as a component segmentation result of the unmanned aerial vehicle infrared photovoltaic image to be segmented according to the component segmentation diagram, the unmanned aerial vehicle infrared photovoltaic image to be segmented and the optimization post-processing.
4. The method for segmenting the component based on the unmanned aerial vehicle infrared photovoltaic image according to claim 3, wherein the step of obtaining the component segmentation map after the optimization post-processing according to the component segmentation map and the unmanned aerial vehicle infrared photovoltaic image to be segmented and the optimization post-processing comprises the following steps: Obtaining a noise pixel ratio and a noise standard deviation of the component segmentation map, and calculating a structure kernel size according to the noise pixel ratio and the noise standard deviation to obtain a structure kernel size parameter; Obtaining a one-dimensional array corresponding to the intensity gradient map and the length of the one-dimensional array, carrying out ascending arrangement on the one-dimensional array, and carrying out position index calculation on the one-dimensional array after ascending arrangement according to the length of the one-dimensional array by adopting a linear interpolation method to obtain a position index; If the position index is not an integer, taking an integer part of the position index as a first sub-position index and a decimal part of the position index as a second sub-position index, and carrying out edge detection threshold calculation according to the first sub-position index, the second sub-position index and the one-dimensional array after the ascending order to obtain a first edge detection threshold; and according to the first edge detection threshold and the second edge detection threshold, carrying out edge detection on the unmanned aerial vehicle infrared photovoltaic image to be segmented to obtain a second intermediate component segmentation map, overlapping the first intermediate component segmentation map and the second intermediate component segmentation map, and obtaining the optimized component segmentation map.
5. Subassembly cutting device based on unmanned aerial vehicle infrared photovoltaic image, its characterized in that includes: the system comprises a data acquisition module, a processing module and a processing module, wherein the data acquisition module is used for acquiring an unmanned aerial vehicle infrared photovoltaic image to be segmented and a preset photovoltaic image segmentation model, and the photovoltaic image segmentation model comprises a feature extraction module, a feature reconstruction module, a feature strengthening module, a component segmentation module and a post-processing module; The multi-scale feature extraction module is used for inputting the unmanned aerial vehicle infrared photovoltaic image to be segmented into the coding module for coding processing to obtain a feature coding diagram; Inputting the characteristic coding diagram into a convolution layer of the multi-scale pyramid pooling module for convolution processing to obtain a convolution characteristic diagram, and pooling the convolution characteristic diagram serving as an input characteristic diagram of a first pooling layer to obtain a characteristic pooling diagram output by the first pooling layer; repeatedly carrying out pooling treatment until the characteristic pooling diagram output by the last pooling layer is obtained, and taking the characteristic pooling diagram as a characteristic extraction diagram to obtain a plurality of scale characteristic extraction diagrams; The sub-pixel structure feature reconstruction module is used for inputting the feature extraction graphs of the scales into the feature reconstruction module, and carrying out up-sampling layer-by-layer fusion on the feature extraction graph of the last scale by adopting a stepwise bilinear interpolation method according to the feature extraction graphs of the scales to obtain feature fusion graphs of the scales; Carrying out sub-pixel convolution on the feature fusion map of a plurality of scales to obtain a sub-pixel convolution feature map of a plurality of scales; the edge feature enhancement module is used for inputting the feature reconstruction into the feature enhancement module, carrying out multi-mode gradient map calculation on the unmanned aerial vehicle infrared photovoltaic image to be segmented by adopting a Sobel operator to obtain an intensity gradient map and a texture gradient map, and fusing the intensity gradient map and the texture gradient map to obtain a composite edge response map; performing attention weight calculation according to the composite edge response graph to obtain attention weight parameters; multiplying the characteristic reconstruction graph by the attention weight parameter element by element to obtain an edge characteristic enhancement graph; And the image component segmentation module is used for inputting the edge characteristic enhancement image into the component segmentation module to carry out component segmentation to obtain a component segmentation image which is used as a component segmentation result of the unmanned aerial vehicle infrared photovoltaic image to be segmented.
6. A computer device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the unmanned aerial vehicle infrared photovoltaic image-based component segmentation method according to any one of claims 1 to 4.
7. A storage medium storing a computer program which, when executed by a processor, implements the steps of the unmanned aerial vehicle infrared photovoltaic image-based component segmentation method according to any one of claims 1 to 4.

Description

Unmanned aerial vehicle infrared photovoltaic image-based component segmentation method and device Technical Field The invention relates to the technical field of computer vision and image processing, in particular to a component segmentation method, device, computer equipment and storage medium based on unmanned aerial vehicle infrared photovoltaic images. Background With the rapid development of the photovoltaic industry, it is important to monitor the photovoltaic panel efficiently and accurately. The unmanned aerial vehicle carries the photovoltaic image that infrared device obtained, can detect the photovoltaic board fast, on a large scale, discovers potential trouble and defect. At present, the deep learning model is used for dividing the photovoltaic module, but the problems of complex background interference, difficult detection of small targets, high data marking requirement, large consumption of computing resources, insufficient generalization capability of the model, influence of illumination change and shielding, insufficient instantaneity, post-processing complexity and the like still exist, accurate module division of the unmanned aerial vehicle infrared photovoltaic image is difficult, and false detection is easy to occur. Disclosure of Invention Based on the above, the invention aims to provide a component segmentation method, device, equipment and storage medium based on unmanned aerial vehicle infrared photovoltaic images, which are combined with sub-pixel structure feature reconstruction and a multi-mode gradient edge attention mechanism to effectively inhibit background noise of the unmanned aerial vehicle infrared photovoltaic images, reduce the background misjudgment rate, improve the accuracy of identifying sub-pixel level structures, thereby improving the component edge segmentation precision of the unmanned aerial vehicle infrared photovoltaic images, effectively overcoming the problems that accurate component segmentation is difficult to be carried out on the unmanned aerial vehicle infrared photovoltaic images under complex illumination and shielding scenes, and easy occurrence of missed detection misdetection, and meeting the industrial detection requirements. In a first aspect, an embodiment of the present application provides a method for dividing an assembly based on an infrared photovoltaic image of an unmanned aerial vehicle, including the following steps: Obtaining an unmanned aerial vehicle infrared photovoltaic image to be segmented and a preset photovoltaic image segmentation model, wherein the photovoltaic image segmentation model comprises a feature extraction module, a feature reconstruction module, a feature reinforcement module, a component segmentation module and a post-processing module; Inputting the unmanned aerial vehicle infrared photovoltaic image to be segmented into the feature extraction module to perform multi-scale feature extraction, and obtaining feature extraction graphs with a plurality of scales; inputting the feature extraction graphs of the scales into the feature reconstruction module to reconstruct the features of the sub-pixel structure, so as to obtain a feature reconstruction graph; Inputting the characteristic reconstruction image into the characteristic strengthening module, and carrying out edge characteristic strengthening according to a multi-mode gradient edge attention mechanism to obtain an edge characteristic strengthening image; Inputting the edge characteristic enhancement map into the component segmentation module to carry out component segmentation, and obtaining a component segmentation map as a component segmentation result of the unmanned aerial vehicle infrared photovoltaic image to be segmented. In a second aspect, an embodiment of the present application provides a component segmentation apparatus based on an unmanned aerial vehicle infrared photovoltaic image, including: the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring an unmanned aerial vehicle infrared photovoltaic image to be segmented and a preset photovoltaic image segmentation model, and the photovoltaic image segmentation model comprises a feature extraction module, a feature reconstruction module, a feature strengthening module, a component segmentation module and a post-processing module; the multi-scale feature extraction module is used for inputting the unmanned aerial vehicle infrared photovoltaic image to be segmented into the feature extraction module for multi-scale feature extraction, and obtaining feature extraction graphs with a plurality of scales; the sub-pixel structure feature reconstruction module is used for inputting the feature extraction graphs of the scales into the feature reconstruction module to reconstruct the sub-pixel structure features to obtain feature reconstruction graphs; the edge feature enhancement module is used for input