The defect detection and impact assessment of photovoltaic arrays are of great significance for improving the performance and reliability of their systems. Due to the problems of missing details and susceptibility to noise interference in the detection images of photovoltaic array hot spot defects, in order to be able to quickly identify defects and accurately analyze their damage, a set of systematic inspection schemes are proposed. Use a handheld thermal imager to take pictures of the photovoltaic array and transfer the acquired images to the computer, and convert the original Red-Green-Blue color mode (RGB) image into grayscale through grayscale processing image. Due to the shortcomings of traditional thermal imaging detection technology such as low efficiency and lack of defect measurement standards, this paper proposes an improved gray conversion algorithm. First, perform K-means clustering of grayscale images, and perform linear enhancement, it can linearly increase the gray value of each area. Then use threshold segmentation algorithm to perform regional correction of image color clusters. Finally, analyze the image quantitatively according to the defect degree table. Data and image combined to show the defects of the photovoltaic array. The experimental results show that compared with the traditional binarization processing algorithm, the deviation of the algorithm proposed in this paper is less than 2% in defect detection, and show the areas of slight defects and severe defects in detail. It also can suppress image non-uniform noise and improve the accuracy of detection.