Abstract:Polycrystalline silicon panels (PSPs) hold immense significance in the photovoltaic industry, with their surface quality profoundly impacting optical performance. Hence, online visual surface defect detection for these panels carries significant implications for the manufacturing process. However, current inspection methods, particularly manual ones, are inefficient. Furthermore, most current automated optical inspection methods heavily rely on large-scale datasets and are insufficient in addressing unseen defect types. To address these challenges, this paper introduces a smart PSP quality inspection system, leveraging cloud–edge computing collaboration. Additionally, we propose a novel Dual Subspace Attention Variational Learning (DS-AVL) approach for few-shot unseen defect segmentation in PSPs. Within the framework of this intelligent inspection system, DS-AVL first extracts foreground and background prototypes and obtains corresponding attention maps using the proposed non-parametric Fore/Background Semantic Separation (FB-SS) module. Then, it performs probabilistic variational inference for the fore/background dual subspace distributions to address the complex texture of PSPs and the inherent uncertainties due to limited samples. Finally, the defective foreground and normal background can be segmented using non-classifier predictions based on statistical properties. The proposed DS-AVL model uniquely integrates attributes from both deep learning and statistical learning, establishing a lightweight yet robust framework that demonstrates substantial interpretability. Furthermore, we have established a publicly available benchmark for the PSP industry, on which we conducted extensive experimentation to demonstrate that our approach not only yields optimal results but also boasts efficient inference performance. More importantly, we found that DS-AVL can be directly adapted to few-shot segmentation tasks of other materials without any fine-tuning.

Polycrystalline silicon wafer defect segmentation based on deep convolutional neural networks

A Wafer Surface Defect Detection Method Built on Generic Object Detection Network

A Sub-Region Unet for Weak Defects Segmentation with Global Information and Mask-Aware Loss.

Polycrystalline Silicon Wafer Scratch Segmentation based on Deep Convolutional Autoencoder

A novel joint segmentation approach for wafer surface defect classification based on blended network structure

A Pixel-Level Segmentation Convolutional Neural Network Based on Global and Local Feature Fusion for Surface Defect Detection

A nondestructive automatic defect detection method with pixelwise segmentation

A Light-Weighted CNN Model for Wafer Structural Defect Detection

Solar Cells Surface Defects Detection Based on Deep Learning

WaferSegClassNet -- A Light-weight Network for Classification and Segmentation of Semiconductor Wafer Defects

Deformable Convolutional Networks for Efficient Mixed-Type Wafer Defect Pattern Recognition

An Adaptive Feature Refinement Network for Pixel-level Segmentation of Surface Defect

AnomalySeg: Deep Learning-Based Fast Anomaly Segmentation Approach for Surface Defect Detection

Efficient Cell Segmentation from Electroluminescent Images of Single-Crystalline Silicon Photovoltaic Modules and Cell-Based Defect Identification Using Deep Learning with Pseudo-Colorization

Real-Time Fabric Defect Segmentation Based on Convolutional Neural Network

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

A High-Efficiency Fully Convolutional Networks for Pixel-Wise Surface Defect Detection

A Deep Convolutional Neural Network for Wafer Defect Identification on an Imbalanced Dataset in Semiconductor Manufacturing Processes

SEMI-PointRend: Improved Semiconductor Wafer Defect Classification and Segmentation as Rendering

Convolution Neural Network Based Polycrystalline Silicon Photovoltaic Cell Linear Defect Diagnosis Using Electroluminescence Images