Solar cell defect detection based on improved G-SSD Network

Aiming at the problems of finger interruptions, microcracks and shadow defects of solar cells that are not easy to detect and will reduce the efficiency of the cells, a solar cell defect localization detection algorithm based on improved G-SSD network is proposed. For object detection algorithms, the detection of small targets is challenging. Firstly, the input scale and network model are improved for the detection of small-scale targets. Secondly, in order to reduce the weight of the network model, a more lightweight network GhostNet is selected instead of VGG-16 in the experiment to reduce the computational cost of the network model. Finally, the EL image is input into the network for classification and positioning, and the prediction results are integrated and the final detection results are output. Experimental results show that compared with the original model, the mean (mAP) of the improved G-SSD is increased by 0.68%, effectively reduces some computing costs.