Online surface defect segmentation on aluminum strip production line using a lightweight and efficient model

Image segmentation is a significant step in image analysis and computer vision. Many entropy-based approaches have been presented on this topic. Among them, Tsallis entropy is one of the best-performing methods. In this paper, the surface defect images of galvanised steel sheets were studied. A two-dimensional asymmetric Tsallis cross-entropy image segmentation algorithm based on chaotic bee colony algorithm optimisation was used to investigate the segmentation of surface defects under complex texture backgrounds. On the basis of Tsallis entropy threshold segmentation, a more concise expression form was used to define the asymmetric Tsallis cross-entropy in order to reduce the calculation complexity of the algorithm. The chaotic algorithm was combined with the artificial bee colony (ABC) algorithm to construct the chaotic bee colony algorithm, so that the optimal threshold of Tsallis entropy could be quickly identified. The experimental results showed that compared with the maximum Shannon entropy algorithm, the calculation time of this algorithm decreased by about 58% and the threshold value increased by about (26%, 54%). Compared with the two-dimensional Tsallis cross-entropy algorithm, the calculation time of this algorithm decreased by about 35% and about 20% was improved in the g-axis direction only. Compared with the two-dimensional asymmetric Tsallis cross-entropy algorithm, the calculation time of this algorithm decreased by about 30% and the threshold values of the two algorithms were almost the same. The algorithm proposed in this paper can rapidly and effectively segment defect targets, making it a more suitable method for detecting surface defects in factories with a rapid production pace.

Surface defects that affect the quality of metals are an important factor. Machine vision systems commonly perform surface inspection, and feature extraction of defects is essential. The rapidity and universality of the algorithm are two crucial issues in actual application. A new method of feature extraction based on contourlet transform and kernel locality preserving projections is proposed to extract sufficient and effective features from metal surface images. Image information at certain direction is important to recognition of defects, and contourlet transform is introduced for its flexible direction setting. Images of metal surfaces are decomposed into multiple directional subbands with contourlet transform. Then features of all subbands are extracted and combined into a high-dimensional feature vector, which is reduced to a low-dimensional feature vector by kernel locality preserving projections. The method is tested with a Brodatz database and two surface defect databases from industrial surface-inspection systems of continuous casting slabs and aluminum strips. Experimental results show that the proposed method performs better than the other three methods in accuracy and efficiency. The total classification rates of surface defects of continuous casting slabs and aluminum strips are up to 93.55% and 92.5%, respectively.

Pose estimation has been a hot topic in the field of machine vision in recent years. Animals exist widely in nature, and the analysis of their shape and movement is important in many fields and industries. In the pose estimation task, to improve the detection accuracy, the existing models often need to consume a lot of computing and memory resources. Therefore, it is a key problem for the pose estimation methods to carry out a lightweight model and reduce the computational overhead on the premise of ensuring model accuracy. In this paper, we focus on the structure of the convolutional neural network in animal pose estimation, construct a lightweight and efficient stacked hourglass network model oriented to optimize the balance of model computation and accuracy, and implement the application algorithm design based on it. Aiming at the problem of large parameters in depthwise convolutional neural networks, a lightweight residual module is proposed, that is, based on the lightweight efficient channel attention improved conditional channel-weighted method (ICCW-Bottle), thereby reducing the weight of the network and obtaining the feature information of different scales. Given the problem that a large amount of feature information is easily lost after the network pooling operation, a lightweight dual-branch fusion module is proposed that fully integrates high-level semantic information and low-level detailed features under the condition of a small number of parameters. Finally, the same as the CC-SSL method: the model is trained jointly using synthetic and real animal datasets, but the CC-SSL method does not take into account the computational power of the model, which consumes a lot of time and memory to run. Through experiments, it is known that compared with the CC-SSL method, the PCK@0.05 of this method is increased by 5.5% on the TigDog dataset. The model in this paper reduces the number of parameters and calculations of the network while ensuring less information loss and model accuracy. The ablation experiment verifies the advancement and effectiveness of the overall network.

In contemporary industrial manufacturing, the automatic detection of surface defects on metal workpieces is a crucial factor for improving product quality and enhancing production efficiency. Therefore, we introduce a novel dual-branch feature interaction network equipped with two advanced attention mechanism specifically designed for defect segmentation, named DFI-Net. Specifically, we adopt an encoder–decoder framework to facilitate effective feature learning for image processing tasks. To enrich feature interaction and improve representation capability, we build a dual-branch architecture that enables complementary information exchange between parallel pathways. Within the encoder stage, we design a bottleneck dilated convolution module and a depthwise-pooling feature mixer module to expand the receptive field and extract multi-scale contextual features. Subsequently, we integrate two specialized modules in the transition layer of the encoder and decoder. The first is a multi-dilation soft attention fusion module, which aggregates features obtained at different dilation rates and dynamically reweights them through soft attention mechanisms. The second is a large kernel attention module, which captures long-range dependencies and broader spatial patterns using expanded convolutional kernels. Finally, the proposed DFI-Net was rigorously evaluated using both the publicly available SD-saliency-900 dataset and a proprietary Bearing surface defect dataset collected from actual industrial scenarios. On the SD-saliency-900 dataset, DFI-Net achieved Dice of 0.8860, Mcc of 0.8701, and Jaccard of 0.7976. Meanwhile, on the Bearing dataset, DFI-Net obtained Dice of 0.7873, Mcc of 0.7884, and Jaccard of 0.6571. Furthermore, on the public road crack dataset DeepCrack, the DFI-Net achieved Dice of 0.8645, Mcc of 0.8612, and Jaccard of 0.7668. Experimental results across these datasets consistently demonstrated that DFI-Net exhibits high segmentation precision, strong robustness, and excellent adaptability to different defect types. In addition, comprehensive ablation studies were carried out to systematically assess the individual contributions of each component within the proposed network architecture.

Defect classification is an important part of automated surface defect detection systems. Generally, the types of surface defects of industrial products are complex and diverse, and the correct defect type classification can help to subsequently extract the characteristic of defects. Moreover, defect classification can help to count the number of defect modes automatically and provide data support for the precise maintenance of product production. Due to the large number of surface defects and the small difference between defect types, traditional classification methods are difficult to classify defect accurately. Therefore, in order to improve the accuracy of defect image classification, this paper proposes a defect image classification method based on transfer learning and sparse coding. Firstly, a deep CNN feature extraction algorithm for defect images is proposed in combination with transfer learning. Then, the deep CNN features of the defect image are dimension-reduced and sparsely optimized using the sparse coding techniques, and the sparse CNN features are obtained. Finally, the sparse CNN features are classified to realize the defect type determination using the linear SVM. The accuracy of the proposed method is verified by using a steel surface defect image benchmark database, and the effectiveness of the proposed method is proved.

A lightweight detector based on attention mechanism for aluminum strip surface defect detection

This paper focused on the light source optimization for visual inspection of piston surface defects. A novel structured lighting system was developed to overcome the difficulties in the well-qualified imaging of tiny and random defects on the cylindrical smooth surface of components such as pistons. A polarized light-filtering based method was proposed to enhance the defect image contrast, enlarge the detection range, and suppress the noise of flawless area. It was both experimentally proven and theoretically proven effective by mathematical modeling of light intensity loss during light propagation. To analyze the influence of the pose and position of light sources on defect imaging quality, the system model for camera, light source, and workpiece was established, and the structure parameters of the lighting system were optimized to improve the brightness uniformity over the entire inspection region and lighting efficiency. Moreover, a soft lighting method was introduced as a subsidiary means of removing the surface noise at two ends of a piston. This method was found to be more effective compared with conventional software filtering algorithm. Finally, a case was examined. The defects on the automotive brake master piston surface were captured using the optimized light source system. A random sampling experiment with 600 specimens revealed that the false-positive rate was about 2.3%, and with no leak detection. The findings have been applied in the production, and it is expected to be referred in inspecting defects on other smooth cylindrical surfaces.

UGC video quality assessment (UGC-VQA) is a challenging research topic due to the high video diversity and limited public UGC quality datasets. State-of-the-art (SOTA) UGC quality models tend to use high complexity models, and rarely discuss the trade-off among complexity, accuracy, and generalizability. We propose a new perspective on UGC-VQA, and show that model complexity may not be critical to the performance, whereas a more diverse dataset is essential to train a better model. We illustrate this by using a light weight model, UVQ-lite, which has higher efficiency and better generalizability (less overfitting) than baseline SOTA models. We also propose a new way to analyze the sufficiency of the training set, by leveraging UVQ’s comprehensive features. Our results motivate a new perspective about the future of UGC-VQA research, which we believe is headed toward more efficient models and more diverse datasets.

Application of Shearlet transform to classification of surface defects for metals

In order to realize automatic surface defect detection for large aperture precision spherical optical elements, an automatic surface defect detection instrument has been designed. The instrument consists of a microscopic imaging system, illumination system, motion scanning system, and a software algorithm system. Firstly, a multi-angle channel illumination source and a coaxial illumination source were designed. Bright and dark field images of surface defects were captured by cooperating with an automatic zoom microscope. Then, algorithms for scanning trajectory planning, image stitching, and intelligent defect recognition were designed to achieve full-aperture surface image acquisition and defect quantification detection. The automated defect detection process of the instrument is summarized and introduced. Finally, the experimental platform was constructed, which can work well for the optical elements with a maximum diameter of 400 mm and a relative aperture R/D value of 1. It takes about 15 min to detect an optical element with a diameter of 200 mm in dark-field imaging mode. As a result, the minimum line width of scratch detectable is 2 μm and the minimum diameter of pitting detectable is 4 μm. Clearly, the instrument can realize the automatic detection of surface defects of spherical optical elements, and has the advantages of a high efficiency, stability, reliability, quantification, and data traceability.

Object detection is a fundamental task in the field of computer vision, and one of its essential requirements is high-quality feature fusion. Previous works have made various efforts in this regard: CNN-based detectors use convolutional blocks to fuse local features and dense prior knowledge to predict objects, while query-based detectors fuse global features by self-attention then decode features with object queries. However, their feature fusion methods are relatively monotonous. Considering that different modules are applicable to different dimensions, we proposed an object detector named CFFMixer which used hybrid architecture to achieve multi-dimensional feature fusion. The sampling strategy to extract abundant local and global features was first introduced then the Comprehensive Feature Fusion Network (CFFN) was proposed to integrate them. CFFN not only achieved local and global features interaction in the spatial dimension, but also fused semantics in the channel dimension. Furthermore, we conducted experiments and made a comparison with competitive models, our model finally got 43.0 mAP on COCO 2017 dataset within 12 epochs. Experimental results showed that the model’s accuracy benefits from the powerful feature fusion capability of CFFN. Besides, we performed ablation studies on our modules to evaluate their effectiveness.

Target detection algorithm takes a central role in autonomous driving vision system, participating in a wide rang of applications in the obstacle detection, vehicle detection, traffic state detection, and others. In recent years, accurate and efficient deep learning models have been proposed for real-time applications. However, the most accurate approaches tend to employ efficient feature fusion method and improved training strategy, such as spatial pyramid pooling. This practice limits real-time performance due to intense extra computational cost. In this paper, we avoid the traditional feature fusion method by proposing an efficient feature fusion method based on feature bi-directional fusion technique and introduce a training method with noise based on confidence loss sequence. We demonstrate the efficiency of the proposal on the PASCAL VOC2007 and KITTI sets, which include the common target detection task and typical autonomous driving target recognition task. The experimental results show that our approach significantly improve the accuracy for complex background samples and reduce the false detection rate of small targets by keeping real-time performance.

Fabric defect segmentation is an important part of ensuring the quality of product production. Using fabrics with surface defects will affect the quality and reputation of their products. In previous studies, some model compression methods have helped semantic segmentation models to be deployed on resource-limited working devices. However, the capacity reduction of models usually leads to a decline in detection performance. We propose a knowledge distillation method combining traditional KD loss and contrastive relational distillation, which makes student models learn the differential representation among various defects while receiving knowledge transfer from the teacher model. We use DeepLabV3+ and PSPNet with MobileNetV2 backbone networks as student models to validate our method. Experiment shows that our method performs better than direct training and traditional knowledge distillation methods on the DAGM and AITEX datasets. Our method enables lightweight models to achieve higher performance on fabric defect segmentation tasks.

We propose a remote sensing image semantic segmentation model based on dual attention and multi-scale feature fusion to solve the problems of objects scale differences and missing small objects. This model uses ResNet50 in the coding part to extract features. First of all, the output features of each stage of ResNet50 are introduced into the pyramid pooling module, making full use of the multi-scale context information of the image to cope with the change of the object scales. Secondly, the dual attention is introduced in the final output features of ResNet50 to establish the semantic relationship between the spatial and channel dimensions, which enhances the ability of feature representation and improve the condition that small targets are difficult to segment. Finally, starting from the output features of the attention module, the features of all levels are gradually integrated to complete decoding to refine the target segmentation edge. The designed comparative experiments results show the effectiveness of the proposed method.

Surface inspection is a necessary process of fabric quality control. However, it remains a challenging task owing to diverse types of defects, various patterns of fabric texture, and application requirements for detection speed. In this article, a lightweight deep learning model is therefore proposed to complete the segmentation of fabric defects. The input of the model is a fabric image, and the output is a binary image. Generally known, a deep learning model usually needs much data to update the parameters. Still, as an abnormal phenomenon, fabric defects are unpredictable, which makes it impossible to collect a large number of data. Distinct from other models, the proposed method is a supervised network but does not need manually labeled samples for training. A fake sample generator is designed to simulate the defect image, which only needs the defect-free fabric image. The proposed model is trained with fake samples and verified with real samples. The experimental results show that the model trained with false data is useful and achieves high segmentation accuracy on real fabric samples. Besides, a loss function is proposed to deal with the problem of imbalance between the number of background pixels and the number of defective pixels in the fabric image. Comprehensive experiments were performed on representative fabric samples to verify the segmentation accuracy and detection speed of this method.