ACA-Net: An Adaptive Convolution and Anchor Network for Metallic Surface Defect Detection

Abstract

Metallic surface defect detection is critical to ensure the quality of industrial products. Recently, human-advanced surface defect detection algorithms have been proposed. Most of these algorithms rely on convolutional neural networks (CNN) and an anchoring scheme. However, a convolution unit only samples the input feature maps at fixed shapes and locations. Similarly, a set of anchors are uniformly predefined with fixed scales and shapes, which increases the difficulties of bounding box regression. Therefore, we propose an adaptive convolution and anchor network for metallic surface defect detection, named ACA-Net. Specifically, an adaptive convolution and anchor (ACA) module is proposed, which mainly consists of adaptive convolution and an adaptive anchor. Firstly, an adaptive convolution module (ACM) is designed, which adaptively determines the location and shape of each convolution unit. In addition, a multi-scale feature adaptive fusion (MFAF) is proposed, which is used in ACM to extract and integrate multi-scale features. Then, an adaptive anchor module (AAM) is proposed to yield more suitable anchor boxes by adaptively adjusting shapes. Extensive experiments on NEU-DET dataset and GC10 dataset validate the performance of the proposed approach. ACA-Net achieves 1.8% on NEU-DET dataset higher Average Precision (AP) than GA-RetinaNet. Furthermore, the proposed ACA module is also adopted in GA-Faster R-CNN, improving the AP by 1.2% on NEU-DET dataset.

Surface defect detection of hot rolled steel based on multi-scale feature fusion and attention mechanism residual block

“…The approach used in this study is to divide the dataset into two groups: a training set and a test set. Refer to the following papers 56 – 58 , the network model is trained on about 70% of the data that are randomly selected, and the accuracy and robustness of the model are tested on the remaining 30% of the data, as shown in Table 1 . Many of the defects in the datasets have relatively modest sizes and diverse irregular shapes and patterns.…”

Deep Learning for Automatic Vision-Based Recognition of Industrial Surface Defects: A Survey

“…Inspired by Guided Anchor, Chen et al [242] leveraged semantic features to yield more suitable anchor boxes for different surface defects. They proposed an Adaptive Anchor Module (AAM) that first insights on locations where surface defects are likely to exist, and then predicts the shapes at different location [230].…”

“…Transfer learning is commonly deployed with supervision and fine-tuning, but there exists unsupervised transfer learning in which source data is labelled and target data is unlabelled. Knowledge is transferred into domain specific applications with usually few categories (than ImageNet dataset [319] with 1000 classes, which is widely adopted in many image-based defect recognition tasks [81], [224], [242], [275], [276], [277], [278]) by importing trained weights as warm or frozen checkpoints in the new backbone. In the first case network re-weights all layers back-propagating the error on the handful target images; in the second case, freezes shallower layers and fine tunes only deeper ones.…”

Defect Detection Algorithm for Battery Cell Casings Based on Dual-Coordinate Attention and Small Object Loss Feedback

“…Currently, in the field of metal surface quality inspection, deep learning methods have become mainstream [4][5][6][7]. Chen et al [8] proposed an adaptive convolution and anchor network for metal surface quality inspection, introducing a multi-scale feature adaptive fusion that effectively extracts and integrates features from different levels and scales, considering both channel and spatial attention, thereby improving the performance of the detector. Zhao et al [9] proposed a semi-supervised, transformer-based multi-scale feature pruning fusion method that can reasonably detect metal surface quality even in cases with limited samples of rust, scratches, and labeled samples.…”