ASOD: Attention-Based Salient Object Detector for Strip Steel Surface Defects

This study aims to enhance the accuracy and robustness of surface defect detection in steel strips. Addressing the challenges of complex defect morphology, low detection accuracy, and high false alarm rates in existing methods, this paper introduces deep learning techniques. It optimizes the YOLOv7 algorithm, proposing multiple improvement strategies. Specifically, these include: employing deformable convolutions (DConv) to address morphological variations in defects; designing a bidirectional cross-attention feature fusion module (BCC) to leverage information across different feature layers fully; constructing a decoupled detection head (TSCODE) to optimize classification and localization tasks; utilizing the SIoU loss function to accelerate bounding box regression; and applying the K-means++ algorithm to obtain predefined anchor boxes. Experimental results demonstrate that the optimized algorithm achieves a mean average precision (mAP) of 80.1% on the strip steel surface defect dataset, representing a 6.0% improvement over the original method. The proposed optimization strategies significantly enhance the performance of strip steel surface defect detection.

Detection of surface defects in strip steel is crucial in industrial automation to ensure steel quality. Recently, learning-based detection methods have gained prominence among researchers for their robust capabilities. However, accurately delineating clear area boundaries remains a challenge, particularly in the complex backgrounds of steel surfaces. In this paper, we propose the edge-aware Interactive Refinement Network (EAIRNet) to address this challenge. EAIRNet comprises three core modules: a Multiscale Feature Extractor, a Sal-Edge Interactive Module, and a regional attention guided module (RAGM). Firstly, our network extracts multiscale features using ResNet from the steel surface environment. Subsequently, features from encoder’s stage 3–5 are integrated to derive edge features. The SEIM facilitates interaction between salient and edge features in the decoder stage, thereby enhancing the network’s focus on defect area boundaries and refining detection precision. Additionally, we introduce the RAGM to improve the focus of shallow decoder layers on defect areas. Our EAIRNet’s effectiveness is proven through comprehensive experimental results on the public dataset named SD-Saliency-900, in comparison to state-of-the-art methods.

Steel is widely used in the aerospace, machinery and automotive industries. Surface defects not only have a negative impact on the appearance of steel but also significantly reduce its wear resistance, high temperature resistance, corrosion resistance and fatigue strength. Therefore, the detection of steel surface defects is very important to improve the quality of steel production. The limited availability of surface defect samples in the industrial sector poses significant challenges for the accurate detection of defects in high-quality materials. In addition, the existing defect detection model is highly complex and not easy to deploy. To solve this problem, a lightweight defect detection network suitable for steel defects is proposed. The cross-layer feature fusion (CFF) in the design enables effective utilisation of multi-layer semantic features, facilitating the detection of small defects in steel. Secondly, a new loss function is designed to make up for the problems of small data volume and uneven data distribution. The experimental results demonstrate that the steel surface defect detection method proposed in this paper achieves the highest detection performance on widely used public datasets such as RSDDS, NEUS and NRSD-CR(test), while maintaining the lowest model complexity.

This paper proposes a steel strip surface defect detection method based on improved MobileNetV2 and semi-supervised learning. Addressing the issues of high computational complexity and insufficient few-shot learning capability in traditional steel strip defect detection approaches, this study introduces Squeeze-and-Excitation (SE) attention mechanisms and a multi-scale Feature Pyramid Network (FPN) to enhance MobileNetV2 architecture. Combined with a consistency regularization-based semi-supervised learning strategy, the proposed method significantly improves detection performance under limited annotated data conditions. Experimental results demonstrate that the approach achieves 97.3% mean accuracy on steel strip surface defect datasets while maintaining high computational efficiency, providing a feasible solution for real-time defect detection in industrial scenarios.

In the context of industrial hot-rolled strip steel surface defect detection, where the demands for real-time performance and classification accuracy are paramount, we present EAF-DenseNet121–a lightweight, enhanced model that incorporates edge-entropy attention mechanisms. At the inception of the DenseNet121 architecture, we incorporate a learnable Sobel-based edge extraction branch, which is designed to adaptively delineate defect contours with precision. We have designed an Entropy-Attention Fusion (EAF) module to further refine the model’s performance. This module constructs a four-dimensional tensor, integrating the primary feature map, edge map, and their corresponding local entropy maps. By applying dual-path channel-wise and spatial attention, we achieve a weighted fusion of information, thereby enriching the feature representation. The EAF module replaces three pivotal convolutional layers within the DenseNet framework–immediately following the initial convolution and subsequent to the first and second Transition layers. This replacement enhances feature representation with a negligible increase in additional parameters, leading to a substantial improvement in defect recognition and classification accuracy. Our experimental results, obtained on the NEU-DET dataset, reveal that the enhanced model achieves a classification accuracy of 99.17%, representing an improvement of 2.78% over the baseline. Furthermore, on the GC10-DET dataset, the model achieves a classification accuracy of 82.89%, further validating its strong generalization capabilities.

An efficient re-parameterization feature pyramid network on YOLOv8 to the detection of steel surface defect

The automatic detection and recognition of steel surface defects is very important in the steel manufacturing industry based on the classification of defect types using artificial intelligence. Deep learning plays an important role to decrease the defection rate in the steel manufacturing industry. Defects in the surface of steel have a serious effect on the quality of steel. Previous research for the NEU-DET dataset has shown that there are some limitations, such as low detection accuracy. In this paper Feature Pyramid Network (FPN) + Resnet neural network has been proposed for detecting the overlapping and multiple defects in the complex background for improving the defect detection rate. The histogram and edge detection approach is used for feature extraction for the classification of defects. The final result shows that the proposed FPN + Resnet model trained with good steel surface defect detection performance, and the dice score is 0.7963, IoU score is 0.6818 and losses is 0.0522 which is very low as compared to others. The proposed detection approach can efficiently detect minor targeted defects on the steel surface, which can be used as a benchmark for automated steel surface defect identification.

With the development of productivity, people set higher demands on the quality of steel. In recent years, artificial intelligence, especially deep learning-based computer vision technology has attracted great attention, which can be used for detecting steel surface defects such as scratches, patches, and rust spots. However, due to the complexity of the strip steel surface, it is still a challenge to accurately and effectively detect defect regions by the existing defects detection methods. Therefore, we propose a unique saliency model, i.e., deeper feature integration network to highlight the defect regions on the strip steel surface. To be specific, after each encoder stage, we introduce the multiscale global feature extraction module to elevate the multiscale deep features from the encoder. Meanwhile, we deploy the deeper feature extraction module, which contains a bidirectional feature extraction unit, to dig the effective representation for defects. Particularly, the forward one is equipped with a channel-space weighted module and the backward one is equipped with a split attention module. After that, the features from the two branches are progressively integrated by the decoder, yielding the final high-quality saliency maps that give a good depiction of the strip steel surface defects. Extensive experiments are executed on the public dataset, and the comparison results show that our model performs better than the 15 state-of-the-art methods, which proves the effectiveness and superiority of the proposed model.

Saliency detection for strip steel surface defects using multiple constraints and improved texture features

The corrosion segmentation technique powered by deep learning enables early detection and timely intervention, effectively preventing further progression and ensuring the safety and extended service life of steel structures. To address the challenges of complex boundaries and fine-grained features in corrosion regions, this study proposes an enhanced high-resolution network version 2 (HRNetV2) model, termed HRNetV2-corrosion, for high-precision corrosion segmentation and analysis. First, an interlayer feature fusion module was designed to improve the integration of high- and low-level features, thereby enhancing fine-grained feature representation. Second, a global feature fusion module is introduced to strengthen the model’s ability to capture global context. To improve computational efficiency, depthwise separable convolution is employed in the output module, significantly reducing complexity and accelerating inference speed. Furthermore, the integration of the channel attention mixer with spatial refinement (CAMixerSR) and coordinate attention (CoordAtt) mechanisms enhances detail extraction and improves the recognition of blurred boundaries. Experimental results demonstrated that HRNetV2-corrosion not only outperforms traditional models across multiple evaluation metrics but also achieves superior efficiency. In addition, a graded evaluation method based on corrosion area is proposed, which classifies corrosion into distinct levels through quantitative analysis of segmentation results, providing a reliable basis for repair and maintenance decisions. Overall, HRNetV2-corrosion offers an effective solution for high-precision corrosion detection and demonstrates substantial practical value.

Surface-defect detection is crucial for assuring the quality of strip-steel manufacturing. Strip-steel surface-defect detection requires defect classification and precision localization, which is a challenge in real-world applications. In this research, we propose an improved PP-YOLOE-m network for detecting strip-steel surface defects. First, data augmentation is performed to avoid the overfitting problem and to improve the model’s capacity for generalization. Secondly, Coordinate Attention is embedded in the CSPRes structure of the backbone network to improve the backbone network’s feature extraction capabilities and obtain more spatial location information. Thirdly, Spatial Pyramid Pooling is specifically replaced for the Atrous Spatial Pyramid Pooling in the neck network, enabling the multi-scale network to broaden its receptive field and gain more information globally. Finally, the SIoU loss function more accurately calculates the regression loss over GIoU. Experimental results show that the improved PP-YOLOE-m network’s AP, AP50, and AP75, respectively, achieved 44.6%, 80.3%, and 45.3% for strip-steel surface defects detection on the NEU-DET dataset and improved by 2.2%, 4.3%, and 4.6% over the PP-YOLOE-m network. Further, our method has fast and real-time detection capabilities and can run at 95 FPS on a single Tesla V100 GPU.

FaNet: Feature-aware network for few shot classification of strip steel surface defects

With the development of industrial economy, more and more enterprises use machine vision and artificial intelligence to replace manual detection. Therefore, the research of steel surface defect detection based on artificial intelligence is of great significance to promote the rapid development of intelligent factory and intelligent manufacturing system. In this paper, Yolov5 deep learning algorithm is used to build a classification model of steel surface defects to realize the classification and detection of steel surface defects. At the same time, on the basis of Yolov5, combined with the attention mechanism, the backbone network is improved to further improve the classification model of steel surface defects. The experiment shows that the Recall and mAP of improved Yolov5 perform better on the steel surface defect data set. Compared with Yolov5, the number of C3CA-Yolov5 parameters decreased by 13.02%, and the size of pt files decreased by 12.72%; the number of C3ECA-Yolov5 parameters decreased by 13.36%, and the size of pt files decreased by 13.22%.

In the context of resource constrained embedded terminals, the current deep learning-based technology for detecting surface defects in strip steel faces challenges with low detection accuracy and efficiency. This study introduces a lightweight convolution neural network (CNN) model for detecting surface defects of strip steel, which is based on deep learning. Through the integration of non-parameter attention mechanism SimAM, inverse residual structure, and depth-wise convolution module, a lightweight attention mobile backbone network is developed to achieve optimal feature extraction. To address the challenge of detecting small surface defects the context augmentation module is introduced to provide more information for small defects detection by using multi-scale features. To improve the efficiency of feature fusion and reduce parameter redundancy, the GVFPN neck network is proposed. The network aims to represent and deal with multi-scale features objectively while minimizing costs. The experimental results indicate that on the NEU-DET dataset, the proposed network attains an mAP of 78.8% and FPS at 97 frame/s. Moreover, the model requires only 2.39 M parameters and 3.1 G FLOPs. Compared to YOLOv5s, the proposed network reduces parameters by 65.9% and FLOPs by 80.7%, while achieving a 1.3% higher mAP and a 30 frame/s increase in detection speed. These results effectively demonstrate the effectiveness and superiority of the proposed method.

Surface defects have a serious detrimental effect on the quality of steel. To address the problems of low efficiency and poor accuracy in the manual inspection process, intelligent detection technology based on machine learning has been gradually applied to the detection of steel surface defects. An improved YOLOv8 steel surface defect detection model called YOLOv8-MGVS is designed to address these challenges. The MLCA mechanism in the C2f module is applied to increase the feature extraction ability in the backbone network. The lightweight GSConv and VovGscsp cross-stage fusion modules are added to the neck network to reduce the loss of semantic information and achieve effective information fusion. The self-attention mechanism is exploited into the detection network to improve the detection ability of small targets. Defect detection experiments were carried out on the NEU-DET dataset. Compared with YOLOv8n from experimental results, the average accuracy, recall rate, and frames per second of the improved model were improved by 5.2%, 10.5%, and 6.4%, respectively, while the number of parameters and computational costs were reduced by 5.8% and 14.8%, respectively. Furthermore, the defect detection generalization experiments on the GC-10 dataset and SDD DET dataset confirmed that the YOLOv8-MGVS model has higher detection accuracy, better lightweight, and speed.