For a typical surface automated visual inspection (AVI) instrument of planar materials, defect classification is an indispensable part after defect detection, which acts as a crucial precondition for achieving the on-line quality inspection of end products. In the industrial environment of manufacturing flat steels, this task is awfully difficult due to diverse defect appearances, ambiguous intraclass and interclass distances. This paper attempts to present a focused but systematic review of the traditional and emerging automated computer-vision-based defect classification methods by investigating approximately 140 studies on three specific flat steel products of con-casting slabs, hot-rolled steel strips and cold-rolled steel strips. According to the natural image processing procedure of defect recognition, the diverse approaches are grouped into five successive parts: image acquisition, image preprocessing, feature extraction, feature selection and defect classifier. Recent literature has been reviewed from an industrial goal-oriented perspective to provide some guidelines for future studies, as well as to recommend suitable methods for boosting the surface quality inspection level of AVI instruments.

In-process hot-rolled strip steel is suffering from some complicated yet unavoidable surface defects due to its harsh production environment. The automated visual inspection on defects consistently faces challenges of interclass similarity, intraclass difference, low contrast, and overlapping issue, which tend to trigger false or missed detections. This article proposes a prior-modulated and semantic-aligned dynamic transformer, called PMSA-DyTr. In this framework, a long short-term self-attention embedded with local convolution is designed for assisting an encoder to eliminate noise ambiguity between defects and backgrounds. Then, a semantic aligner is cleverly bridged between the encoder and the decoder to align the sematic for speeding up the convergence, and prior-modulated cross attention is proposed to alleviate the deficiency of samples for a data-driven transformer. Furthermore, a gate controller is innovatively constructed to dynamically select the minimal number of encoder blocks while preserving detection accuracy. The proposed PMSA-DyTr outperforms 19 state-of-the-art models on mean average precision with an inference time of 54.67 ms and visually performs best in detecting low-contrast and multiple small defects.

The morphologies of various surface defects on strip steel suffer from oil stain, water drops, steel textures, and erratic illumination. It is still challenging to recognize defect boundary precisely from cluttered backgrounds. This article emphasizes such a fact that skip connections between encoder and decoder are not equally effective, attempts to adaptively allocate the aggregation weights that represent differentiated information entropy values in channelwise, by importing a stack of cross-attention transformer (CAT) into the encoder–decoder network (EDNet). Besides, a cross-attention refinement module (CARM) is constructed closely after the decoder to further optimize the coarse saliency maps. This newly nominated CAT-EDNet can well address the semantic gap issue among the multiscale features for its multihead attention structure. The CAT-EDNet performs best on insuring defect integrity and maintaining defect boundary details when compared with 12 state-of-the-arts, and the detection efficiency is at 28 fps even under the noise interfered scenario.

Accurate low-contrast defect detection has become a common bottleneck to further improve the performance of automated visual inspection (AVI) instruments. Inspired by visual crypsis, a novel concept of camouflaged defect has been proposed to assist surface defect detection, and then, a camouflaged defect detection network (CDDNet) was proposed. To be specific, a new inception dynamic texture enhanced module (IDTEM) was proposed to aggressively strengthen the indefinable boundaries and deceptive textures. To further explore spatial information over long distance, a lightweight recurrent decoupled fully connected attention (RDFCA) is designed with cost-effective computation. Finally, a new adaptive scale-equalizing pyramid convolution (ASEPC) was designed to achieve cross-scale feature fusion by exploiting the inter-layer feature correlation. The proposed CDDNet obtained competitive mean average precision (mAP) of 84.2%, 96.7%, and 67.1%, respectively, on three public datasets of NEU-DET, DAGM, and CAMO, when compared with state-of-the-arts.

Visual surface defect detection is crucial for product quality control in the large-scale wood manufacturing industry. This study focuses on how to assist deep learning model in surviving in the challenges brought by complex texture backgrounds. A novel visual defect detection model, inter-layer information guidance feedback networks (I ² GF-Net), is proposed in this paper. To be specific, a top-down feedback encoder (TDFE) is proposed to guide the attention of the low-level feature map, enabling it to focus on the defect regions by incorporating enhanced high-level semantic information. This significantly reduces false positives triggered by intense textures. Meanwhile, a semantic feature texture enhancement (SFTE) method is designed to compensate for high-level semantic features with fine-grained local information, thereby avoiding frequently missed detections resulted from multiple down-sampling in deep models. Furthermore, we provide an option of dual-round feature refinement (DRFR) to pursue a higher mAP in scenarios where sacrificing a certain amount of time is acceptable. Experimental results demonstrate the I ² GF-Net outperforms 13 state-of-the-arts on two benchmark datasets (VSB-DET and NEU-DET), as well as our newly opened wood dataset (OULU-DET), which will be publicly available at http://www.ilove-cv.com/oulu-wood/.

Current deep learning methods for strip steel defect detection prioritize accuracy and computational efficiency but struggle to generalize to new production environments with unseen defect types. While visual foundation models like segment anything model (SAM) offer broader generalization through pretrained visual knowledge, they face significant domain shifts when applied to industrial defect imagery. To address this, we propose AdaptedSAM, a parameter-efficient few-shot fine-tuning framework for SAM that optimizes adapter placement and feature interaction. We first demonstrate that adapter positioning within the encoder critically impacts few-shot adaptation performance. Building on this, we introduce an efficient global optimizer that dynamically allocates adapter weights via cross-block communication across transformer layers. Furthermore, we develop an adapted mask decoder to enhance detailed mask information. Our adaptedSAM achieves state-of-the-art defect detection with precise boundaries in five-shot learning, outperforming 12 state of the art (SOTA) models on the SD-saliency-900 steel dataset while ensuring computational efficiency. Cross-domain evaluations on six benchmarks (industrial defects, remote sensing, medical imaging, natural scenes) demonstrate robust generalization.