An efficient solar panel crack detection system using a novel solcnn model and crack segmentation algorithm

• SOLCNN classifies cracked/non-cracked solar panels with 98.95% average detection accuracy. • Cracked panel detection rate: 99.3%; non-cracked panel detection rate: 98.6%. • Image enhancement boosts Crack Sensitivity, Specificity, and Accuracy to over 98%. • Image enhancement improved crack detection accuracy from 96.07% to 98.7%. This study introduces an automated framework for solar panel crack detection based on a novel Solar Convolutional Neural Network (SOLCNN) integrated with a Cracked Region Segmentation Algorithm (CRSA). Unlike conventional CNN-based approaches that are limited to classification, the proposed framework unifies preprocessing, data augmentation, feature extraction, classification, and segmentation into a cohesive pipeline. The architecture extends the traditional LeNet model with deeper convolutional–pooling layers and an Internal Feature Matrix (IFM), enabling the capture of micro-level cracks and subtle structural variations. In the preprocessing stage, Contrast Limited Adaptive Histogram Equalization (CLAHE) enhances image clarity, while targeted augmentation improves dataset diversity and training robustness. Experimental evaluations demonstrate that SOLCNN achieves an average detection accuracy of 98.95%, surpassing established CNN architectures including LeNet, AlexNet, and VGG. The integration of CRSA enables precise localization of cracked regions, facilitating detailed defect characterization and predictive maintenance. The system further exhibits resilience to environmental variations and dust interference, while maintaining low detection latency suitable for real-time inspection. Practical applications encompass drone-assisted solar farm monitoring, automated defect detection, and large-scale photovoltaic (PV) maintenance, highlighting the potential of the SOLCNN–CRSA framework to enhance inspection efficiency, reduce operational costs, and extend the service life of PV systems.