融合多模态特征与改进YOLOv12的光伏电池板组件缺陷检测OA

Due to the large number of solar panel components in photovoltaic power plants and prolonged exposure to complex and harsh outdoor environments,various defects are prone to occur,posing great challenges to daily maintenance work.To improve the accuracy and efficiency of defect detection in photovoltaic panel components,a photovoltaic panel component defect intelligent detection method that integrates multimodal features and improves YOLOv12 is proposed by utilizing the feature differences of various defects in visible light and infrared images.Firstly,a sub-patch orthogonal mod-ulation module is designed on the YOLOv12 network framework to extract visible light image features,and the percep-tion of weak texture defects is enhanced by constructing a low rank subspace for direction discrimination,thereby alleviating the problem of modal feature expression differences.Secondly,a thermal gradient residual network based on ResNet-50 is introduced to model the temperature change structure in infrared images,enhancing the modeling ability for thermal anomalies and composite defects.Finally,a dual-spectrum coherent focal loss function is designed to explicitly align the feature focus regions under different modalities,improving the modality collaborative perception effect and robustness in complex environments.The experimental results show that the designed improvement strategy has independent effective-ness and collaborative gain ability in enhancing the detection performance of the model.The impoved model can accurately and efficiently identify various types of composite defects on photovoltaic panel components,with detection accuracy,recall rate,and mAP reaching 97.8%,95.4%,and 96.8%,respectively,improved by 4.3,4.6,and 4.4 percentage points,com-pared with YOLOv12,effectively suppressing the occurrence of false positives and false negatives,and maintaining a high inference speed,providing reliable technical support for the intelligent maintenance of photovoltaic power plants.