Abstract:Defect detection in the industry is an essential task in quality inspection. The main target is to classify and localize defects in acquired images. During image acquisition, external noise and diverse background patterns can lead to conflicting information during fusion of network features, which brings certain challenges to detection. In addition, one-stage detectors generate mismatches when performing classification and localization, which can lead to a certain degree of misalignment in prediction. Also, some small defects to be detected in industrial products must be combined with fine-grained information. Based on this, we present a novel efficient conflict-filtered network (ECF-Net) to improve the detection of small defects in this article. Our ECF-Net reduces the interference caused by conflicting information in feature fusion. Moreover, the detection branch can combine richer features so that the essential information can be well-obtained. First, we build the efficient feature pyramid network (EFPN) by upward fusing the bottom layer's rich location and texture information. Next, we introduce the noise reduction-residual shrinkage building unit with channelwise thresholds (NR-RSBU-CW) in the skip connection of the same layer to filter noise and irrelevant background information. Finally, in the head part, our proposed regression adjustment predictor-head (RAP-Head) aligns the localization information of all the prediction anchor boxes using the spatial offset maps and then adaptively filters the prediction boxes. Not only is the localization information adjusted effectively but also the defective targets between different scales are also matched to higher quality prediction boxes. Experimental results based on three real industrial datasets, including Aliyun Tianchi fabric dataset (50.8% mAP), PCB dataset (97.6% mAP), and Steel dataset (76.8% mAP), demonstrate that the proposed ECF-N- t achieves competitive results compared with other state-of-the-art (SOTA) methods.

AEGLR-Net: Attention Enhanced Global-Local Refined Network for Accurate Detection of Car Body Surface Defects

AEGLR-Net

GRA-Net: Global Receptive Attention Network for Surface Defect Detection

PGA-Net: Pyramid Feature Fusion and Global Context Attention Network for Automated Surface Defect Detection

AP-Net: a Metallic Surface Defect Detection Approach with Lightweight Adaptive Attention and Enhanced Feature Pyramid

CDDNet: Camouflaged Defect Detection Network for Steel Surface

Efficient Conflict-Filtered Network for Defect Detection

Improved RetinaNet-Based Defect Detection for Engine Parts

A Lightweight Deep Learning Network with an Optimized Attention Module for Aluminum Surface Defect Detection

LFRNet: Localizing, Focus, and Refinement Network for Salient Object Detection of Surface Defects

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

Steel surface defect detection based on sparse global attention transformer

LHA-Net: A Lightweight and High-accuracy Network for Road Surface Defect Detection

LAACNet: Lightweight adaptive activation convolution network-based defect detection on polished metal surfaces

Cascaded adaptive global localisation network for steel defect detection

An Industrial Defect Detection Network with Fine-Grained Supervision and Adaptive Contrast Enhancement.

Normal Reference Attention and Defective Feature Perception Network for Surface Defect Detection

AFFNet: An Attention-Based Feature-Fused Network for Surface Defect Segmentation

EMRA-Net: A Pixel-Wise Network Fusing Local and Global Features for Tiny and Low-Contrast Surface Defect Detection

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