Solar Panel Defect Detection Based on Improved Lightweight YOLOv8n Model

Aiming at the problems of low accuracy, high complexity and poor real-time performance of solar panel defect detection models, a lightweight detection model based on YOLOv8n is proposed. Firstly, the C2f-Faster module is employed to fuse FasterNet Block for extracting crucial feature information while reducing redundant computation through feature streamlining. Secondly, an efficient multi-scale attention mechanism (EMA) is introduced to focus on the extraction of features related to minor defects in the neck layer. Lastly, wise intersection over union (WIoU) loss funtion is incorporated to enhance the model's capabilities in target localization and category classification. Experimental results show that on the public solar panel defect dataset, compared with the benchmark model YOLOv8n, the improved model has the [email protected] increased by 2.3%, the parameter amount reduced by 28.9%and the inference speed increased by 152 frames /s. For the defects of solar panels, the improved model effectively realizes the bidirectional optimization of detection accuracy and lightweight.