Abstract:Visual surface defect inspection provides an important tool for product quality assessment in a wide range of industrial applications. In recent years, diverse convolutional neural network (CNN) models have been devel-oped for the high-accuracy image classification of natural objects. However, it remains a challenging task to develop deep-learning-based approaches for surface defect inspection of industrial products. The main challenge is that surface defects of industrial products occur in low probability, therefore it is impractical to build a large image dataset containing different types of defects to train/fine-tune deep CNN models to achieve satisfactory performance when applied to industrial inspection tasks. To overcome the above-mentioned problem, we build a light-weight defect classification model based on the pre-trained SqueezeNet architecture and present three effective techniques to achieve high-accuracy defect classification using very few collected defective training samples. First, we experimentally demonstrate that it is reasonable to freeze shallower convolutional layers and only fine-tune deeper layers in the last convolutional stage when there are not enough training samples. Second, we integrate batch-size independent Group Normalization (GN) into SqueezeNet to compute stable statistics based on a limited number of defective samples for training the classification model. Third, we add an auxiliary task to perform the similarity comparison of two defective samples during the training phase, significantly increasing the size of the training data and generating more distinctive defect-specific representations. The experimental results on the NEU-CLS dataset and USB-FS dataset verify the effectiveness of the proposed method. Our method can obtain accurate defect classification results when very few defective training samples are provided, i.e., about 97.69% and 82.92% average classification accuracy on the NEU-CLS and USB-FS datasets, respectively, under the 5-shot setting (5 images per defect category).

Eswin-Unet: A Collaborative Model for Industrial Surface Defect Detection

Enhanced Concrete Crack Detection and Proactive Safety Warning Based on I-St-Unet Model

Detection of Defects on Metal Surfaces Based on PoolFormer Auto-Encoder

A Deep Learning-Based Surface Defect Inspection System Using Multiscale and Channel-Compressed Features

Swin Transformer Combined with Convolution Neural Network for Surface Defect Detection

Cas-VSwin transformer: A variant swin transformer for surface-defect detection

An Effective Industrial Defect Classification Method under the Few-Shot Setting Via Two-Stream Training

Surface defect detection and classification of steel using an efficient Swin Transformer

A CAM-based Weakly Supervised Method for Surface Defect Inspection

A Machine Vision-based Fabric Defect Detection Solution for Textile Production Industry Using Object Detection

Efficient surface defect detection using self-supervised learning strategy and segmentation network

CINFormer: Transformer network with multi-stage CNN feature injection for surface defect segmentation

Swin-RGC: Swin-Transformer with Recursive Gated Convolution for substation equipment non-rigid defect detection

Fully Convolutional Networks for Surface Defect Inspection in Industrial Environment

ETDNet: Efficient Transformer-Based Detection Network for Surface Defect Detection

Defect-aware transformer network for intelligent visual surface defect detection

Swin-Transformer -YOLOv5 for lightweight hot-rolled steel strips surface defect detection algorithm

SCL-VI: Self-supervised Context Learning for Visual Inspection of Industrial Defects

Surface defect detection of industrial components based on vision

Unsupervised Anomaly Detection for Surface Defects with Dual-Siamese Network

An Efficient End-to-End Multitask Network Architecture for Defect Inspection