基于语义增强和尺度感知的光伏组件红外图像缺陷检测OA

Aiming at the problems of strong background interference,large scale differences of targets,and difficulty in distinguishing small defects in infrared defect images of photovoltaic modules,leading to low detection accuracy,this paper proposes an infrared image defect detection algorithm for photovoltaic modules based on SESPNet.Firstly,a semantic information enhancement module is constructed and embedded in the backbone network to fuse global and local semantic information,enhance feature expression ability and suppress the interference of complex background noise.Secondly,a spatial attention pyramid pooling module is adopted to replace the original spatial pyramid pooling module in YOLOv10,and the multi-scale defect perception ability is enhanced through the weighted fusion of local and global feature information.Finally,a multi-scale channel attention mechanism is constructed in the neck network to further improve the extraction ability of small-scale feature information by establishing information interaction between different channels.The experimental results based on the self-made infrared defect dataset of photovoltaic modules show that the average value of the average accuracy of SESPNet reaches 92.1％,and the detection speed reaches 62.4 frames/s,which is significantly better than other mainstream detection algorithms.The comparative experiments in the embedded environment prove that SESPNet still has excellent real-time performance and detection performance under limited computing resources.