To address the challenges faced by operators in detecting anomalies in photovoltaic panels under real-world conditions, an image detection algorithm based on YOLOv10n for photovoltaic stations is proposed. This algorithm introduces three key improvements: optimizing the original C2f Bottleneck convolution to enhance multi-scale compatibility and improve the backbone's feature extraction capability; further optimizing the Neck structure to enable multi-level cross-scale connections of the backbone feature maps; and to achieve a more lightweight model, the convolution in the Neck was replaced with GSConv, reducing model redundancy while maintaining efficiency. From the final visualization results, the improved CEG-YOLOv10n model effectively avoids false detections and missed detections while maintaining high recognition accuracy. This provides a new solution for researchers focusing on photovoltaic power station image detection algorithms, contributing to the advancement of renewable energy and the development of photovoltaic power generation. Code and data are available at https://github.com/wljy2022/Solar-panel-defect-image-dataset.git .