Adaptively Fused Attention Module for the Fabric Defect Detection

Abstract

In the fabric manufacturing industry, fabric defect detection is a practical yet challenging task, due to the problem of defects with small sizes or unremarkable appearances distributed in fabric images with high resolution. Some deep‐learning‐based solutions try to tackle the aforementioned problem but with limited achievements. Herein, a brand new module called adaptively fused attention module (AFAM) is proposed to improve the detection performance by enabling the network to concentrate more on the small or unremarkable defects in terms of 1) enhancing the feature maps both spatial‐wise and channel‐wise, 2) enhancing the attention information flow between the channel‐attention feature map and the spatial‐attention feature map, and 3) enriching background information. Experiments on the MS‐COCO dataset show that AFAM reaches higher AP than other state‐of‐the‐art (SOTA) attention modules (e.g., squeeze‐and‐excitation network [SE], efficient channel attention [ECA], convolutional block attention module [CBAM]). Meanwhile, on the Smart Diagnosis of Cloth Flaw Dataset (CloF), the Cascade R‐CNN (Resnext101) + AFAM outperforms the SOTA in the official leaderboard of the CloF dataset. Speed‐wisely, the solution achieves a 12.19‐FPS on RTX 3090, thus making it practical to be applied in the industry of fabric defect detection.

PCB Soldering Defect Inspection Using Multitask Learning under Low Data Regimes

“…There are however not many open-source datasets related to PCB images, and all of them are not about PCB soldering defects. [15][16][17][18][19][20][21][22][23] For this reason, a new PCB soldering defect dataset, namely, PCBSPDefect is constructed. [24] The dataset contains images of PCB soldering points of three component parts, namely, dual in-line packages (DIP) located on the PCB back side (BDIP), DIP on the PCB front side (FDIP), and the flat flexible cables (FFC), as depicted in Figure 1.…”