Few-shot unseen defect segmentation for polycrystalline silicon panels with an interpretable dual subspace attention variational learning framework

Haiming Yao, Wei Luo, Wenyong Yu, Xiaotian Zhang, Zhenfeng Qiang, Donghao Luo, Hui Shi

Abstract: Highlights•An edge–cloud collaborative inspection framework for defect detection in polycrystalline silicon panels (PSPs) is investigated.•We proposed the Dual Subspace Attention Variational Learning (DS-AVL), a novel few-shot defect segmentation method for PSPs.•Presenting PSP-DS, a benchmark dataset designed specifically for defect detection within the photovoltaic industry.