Research on attention weighted enhanced defect recognition algorithm for photovoltaic panel

Abstract

Detecting defects on the surface of photovoltaic (PV) panels is crucial during both the production and operational phases, as surface structural damage is often indicative of potential issues within the internal structure. This paper presents a preliminary screening algorithm for photovoltaic panel defects using optical cameras, aiming for cost-effective and efficient detection. However, images captured by optical cameras face challenges such as small defect sizes and complex backgrounds. Therefore, the detection model needs strong capabilities in detecting small targets and effective feature extraction. Therefore, this paper propose a multipath feature-weighted fu-sion method based on YOLOv7 to enhance high-quality feature paths by leveraging the channel dimension sensitivity of the FCA module. The backbone and feature fusion structure of the original model is reconstructed. Additionally, to address the singularity in the fusion data of the medium-sized output tensor, long residual edges connecting the backbone are incorporated. Fi-nally, experiments conducted on two datasets demonstrate that the enhanced model outperforms YOLOv7, with improvements in [email protected] of 3% and 5.4%, respectively. Meanwhile, the en-hanced model exhibits superior robustness and better balance compared with other mainstream models.