Online PCB Defect Detector On A New PCB Defect Dataset

TL;DR

This work addresses PCB defect detection by introducing a deep detector that operates on aligned template–tested image pairs and a novel Group Pyramid Pooling module to handle multi-scale defects. It introduces the DeepPCB dataset, comprising 1,500 aligned image pairs with six defect types, and demonstrates state-of-the-art detection performance with real-time inference. The SSD-style multi-scale predictions, together with GPP, yield higher accuracy than traditional image-processing baselines and comparable deep detectors, while remaining efficient. The DeepPCB dataset is publicly released to spur further research in PCB defect localization and classification.

Abstract

Previous works for PCB defect detection based on image difference and image processing techniques have already achieved promising performance. However, they sometimes fall short because of the unaccounted defect patterns or over-sensitivity about some hyper-parameters. In this work, we design a deep model that accurately detects PCB defects from an input pair of a detect-free template and a defective tested image. A novel group pyramid pooling module is proposed to efficiently extract features of a large range of resolutions, which are merged by group to predict PCB defect of corresponding scales. To train the deep model, a dataset is established, namely DeepPCB, which contains 1,500 image pairs with annotations including positions of 6 common types of PCB defects. Experiment results validate the effectiveness and efficiency of the proposed model by achieving $98.6\%$ mAP @ 62 FPS on DeepPCB dataset. This dataset is now available at: https://github.com/tangsanli5201/DeepPCB.

Figures (4)

• Figure 1: This figure shows the pair of (a) a defect-free template image and (b) a defective tested image with annotations of the positions and types of PCB defects in the DeepPCB dataset.
• Figure 2: Defect number of the 6 categories in DeepPCB train/validation and test set.
• Figure 3: An overview of the proposed model. The backbone can be any efficient convolutional model, e.g., VGG16-tiny VGG or ResNet18 ResNet without the last fully connected layers. We examine the average pooling and max pooling as the 'Pool' operation in GPP module in the experiments. 'BN' is abbreviation of batch normalization. Up-sample is implemented by bilinear interpolation and the target size is the same as the first input of each concatenated group. Each group in GPP module makes predictions in different scales.
• Figure 4: Comparison of performance and detection speed of the PCB defect detection methods on DeepPCB dataset.