Intelligent Detection System for Dirty Photovoltaic Panels Based on YOLOv8

This article proposes an intelligent detection system for photovoltaic panel contamination based on YOLOv8n, named, which establishes a six-level classification standard to address the issue of quantifying contamination levels. The study uses a dataset of 3,862 photovoltaic panel contamination images, divided into training, validation, and test sets in a ratio of 7:2:1, and trains the model end-to-end using a YOLOv8n model. The network structure is optimized with the C2f module, resulting in a of 3.2M, which is 18% smaller than YOLOv5n, balancing accuracy and lightweight performance. Experiments were conducted under the Ubuntu20.04 environment, using a RTX 4060 GPU, and after 300 training cycles, the model achieved an [email protected] of 0.832, with the highest AP value for the “Very Dirty” category (0.890) and the lowest for the “Clean” category (0.779), demonstrating the model's capability to detect different contamination levels. A human-machine interface based on PyQt has been developed, enabling real-time visualization of detection results and alarm linkage functions. This study provides an efficient solution for the automatic detection of photovoltaic panel dirt and the scheduling problem of cleaning robot, which helps to improve the operation efficiency and reliability of photovoltaic system.