Research on segmentation algorithm of rooftop distributed PV arrays based on deep learning

Abstract

Hot spot detection is a very important aspect in the field of PV plant inspection, and with the development of UAV technology, PV hot spot detection by UAV based on image processing has gradually emerged. To solve this problem, this paper introduces a deep learning model for coarse detection of PV regions to remove background interference, and uses image preprocessing, convolutional operations, morphological operations and Hough line transformation to finally achieve component-level segmentation of PV arrays. The experimental results show that the algorithm of this paper can segment the PV array accurately and quickly, and the effect is better than the traditional segmentation algorithm.