CD-YOLO: A lightweight end-to-end detection model for cigarette appearance defects

Multi-scale defects in powder-based additively manufactured metals and alloys

With the development of deep fusion intelligent control technology and the application of low-carbon energy, the number of renewable energy sources connected to the distribution grid has been increasing year by year, gradually replacing traditional distribution grids with active distribution grids. In addition, as an important component of the distribution grid, substations have a complex internal environment and numerous devices. The problems of untimely defect detection and slow response during intelligent inspections are particularly prominent, posing risks and challenges to the safe and stable operation of active distribution grids. To address these issues, this paper proposes a high-performance and lightweight substation defect detection model called YOLO-Substation-large (YOLO-SS-large) based on YOLOv5m. The model improves lightweight performance based upon the FasterNet network structure and obtains the F-YOLOv5m model. Furthermore, in order to enhance the detection performance of the model for small object defects in substations, the normalized Wasserstein distance (NWD) and complete intersection over union (CIoU) loss functions are weighted and fused to design a novel loss function called NWD-CIoU. Lastly, based on the improved model mentioned above, the dynamic head module is introduced to unify the scale-aware, spatial-aware, and task-aware attention of the object detection heads of the model. Compared to the YOLOv5m model, the YOLO-SS-Large model achieves an average precision improvement of 0.3%, FPS enhancement of 43.5%, and parameter reduction of 41.0%. This improved model demonstrates significantly enhanced comprehensive performance, better meeting the requirements of the speed and precision for substation defect detection, and plays an important role in promoting the informatization and intelligent construction of active distribution grids.

In recent years, detecting defects in materials has become an important element in many industries, such as automotive, construction, textile manufacturing, and many others. Material defects, such as stains, cracks, dents, and others, can affect the quality and durability of the product, which can have serious consequences for safety, performance, and product quality. Therefore, it is important to detect and remove material defects as quickly and accurately as possible. Traditional methods of detecting material defects, such as manual inspections or simple visual algorithms, can be unreliable and, above all, time-consuming. In recent years, however, new tools based on intelligent algorithms (Liu et al., 2023) or structures (Neethu et al., 2015), such as neural networks, have emerged that enable automation and significantly increase the precision of defect detection. One of the most popular tools for object detection is the YOLOv5 (Redmon et al., 2016) (You Only Look Once version 5) model, which is based on neural networks and designed for fast and accurate object detection in images. The YOLOv5 model can be trained to detect specific types of material defects, such as stains, cracks, dents, etc. To use the YOLOv5 model for defect detection in materials, images of the material with specified objects of interest are needed, which will be used as input data for the model training. The YOLOv5 model analyses the images and determines rectangular frames around each defect indicating its location, and then assigns a selected label (in this case, the name of the defect). As a result, the YOLOv5 model is able to detect defects in new material images, regardless of their location, colour, and size. As the work progresses, it is also possible to prepare new images and retrain the model for new defects or to improve the performance of existing ones. As a result of the research, the obtained results at a 95% level of recognized defects certainly qualify the applied technology for professional use after appropriate workstation preparation and a sufficiently large amount of data (preferably 1000 or more).

As an important part of the bridge structure system, the underwater pile-pier structure usually occurs various defects on its surfaces due to its complex hydrological environment. The existing conventional defect detection approaches exist two aspects of problems: (1) insufficient definition and color distortion of the underwater images, and (2) low efficiency and error-prone. To solve these problems, this paper proposed the target defect detection model by integrating the image-fusion enhancement algorithm and the deep learning algorithm. Firstly, by analyzing the reasons for the degradation of the underwater images, the ACE (automatic color equalization) and CLAHE (contrast limited adaptive histogram equalization) algorithms are selected to enhance the image, respectively. Secondly, the two enhanced images are fused based on the point sharpness weight, and then the fusion results are further sharpened by the USM (unsharp mask) algorithm, thus obtaining the final fused images. Thirdly, 3,200 fused images are taken as the training set, by adopting the YOLOv3 algorithm to train the detection model, and then the training model is validated and tested by the other each 400 fused images, thus building up the target automatic detection model of underwater pile-pier surface defects. Finally, a series of comparison and discussion were conducted to validate the effectiveness of image-fusion and the robustness and effectiveness of the target detection model. The results found that the target detection model has excellent robustness against noise and effectiveness in the surface defect detection. This indicates that the image-fusion approach proposed in this paper can effectively enhance the image features, and the target detection model is feasible, robust, and effective in the automatic detection of surface defects on underwater pile-pier structures.

Edge computing can avoid the long-distance transmission of massive data and problems with large-scale centralized processing. Hence, defect identification for insulators with object detection models based on deep learning is gradually shifting from cloud servers to edge computing devices. Therefore, we propose a detection model for insulators and defects designed to deploy on edge computing devices. The proposed model is improved on the basis of YOLOv4-tiny, which is suitable for edge computing devices, and the detection accuracy of the model is improved on the premise of maintaining a high detection speed. First, in the neck network, the inverted residual module is introduced to perform feature fusion to improve the positioning ability of the insulators. Then, a high-resolution detection output head is added to the original model to enhance its ability to detect defects. Finally, the prediction boxes are post-processed to incorporate split object boxes for large-scale insulators. In an experimental evaluation, the proposed model achieved an mAP of 96.22% with a detection speed of 10.398 frames per second (FPS) on an edge computing device, which basically meets the requirements of insulator and defect detection scenarios in edge computing devices.

The accurate detection of steel surface defects remains challenging because of their irregular shapes and complex backgrounds, which often result in missed detections and false positives. Moreover, existing models are unsuitable for edge devices due to large parameters and high computational demands. To address these issues, this paper presents DCDF-YOLO, a lightweight steel surface defect detection model based on YOLOv8n. First, a novel CSPDC feature extraction module replaces the standard C2f module by incorporating dual convolution. Group convolution techniques arrange filters efficiently to optimize information flow and enhance extraction efficiency and representation capacity. Second, a lightweight cross scale feature fusion module named CCFM is introduced during fusion to reduce parameters and computational cost while improving adaptability to scale variations. Third, a Dynamic Deformable Head (DDH) is proposed to improve detection of small defects and integrate feature diversity across scales. This detection head addresses limitations in handling long range dependencies and spatially adaptive aggregation, capturing local details and structural features effectively. Finally, a novel bounding box loss function Focaler-SIoU is introduced. It focuses on regression samples of varying difficulty and incorporates an angular penalty mechanism to enhance precision, inference capability, and robustness in defect recognition. The experimental results demonstrate that the improved model achieves mAP@0.5 gains of 4.5% and 2.7% on the public steel datasets GC10-DET and NEU-DET, respectively, compared to the baseline YOLOv8n. Additionally, the model’s parameter is reduced by 28.6% to 2.15M. Compared with other mainstream object detection models, the DCDF-YOLO model achieves an optimal balance between detection accuracy and lightweight design, meeting the requirements of edge devices operating under limited computational resources.

Hot-rolled steel strip plays an important role in the field of industrial manufacturing. In addition, defects on its surface affect the aesthetics of the subsequent products and their corrosion resistance, wear resistance, and fatigue strength. However, the existing methods are difficult to learn or capture discriminative feature representations, resulting in poor detection performance. Therefore, its surface defect detection faces two main challenges: one is the insufficient ability to extract local features, and the other is the limited ability to detect multi-scale targets. To address the above issues, we propose a Residual Deformable Convolution and Double LayerNorm Swin Transformer and Channel Expansion Feature Pyramid Networks (RTCN) multi-scale hot-rolled strip surface defect detection model, which adopts Double LayerNorm Swin Transformer (DLST) and as Residual Deformable Convolution Block (RDCB) its backbone network to increase the sensitivity of the model's detection of small and irregular defects. In addition, we adopt Channel Expansion Feature Pyramid Networks (CEFPN) to introduce more feature dimensions to better capture the structure and semantic image information. Ultimately, we assess the proposed model using the publicly available NEU-DET dataset. Our comprehensive testing shows that the model developed in this paper beats the most advanced approach by 1.1 % to 7.2 % in mAP.

The quality of welding is directly related to the performance and life of the welding structure. The nondestructive testing technology is mainly used for welding defect defection, while the X-ray testing technology can directly and reliably reflect the shape, location and size of defects. For X-ray images, manual inspection is used at present, and it is easily caused by subjective factors, such as professional level, which may lead to low accuracy. This paper focuses on establishing an end-to-end automatic detection model of X-ray welding defects to improve the accuracy and efficiency of detection based on a deep learning algorithm. Considering the feature information of welding defects, this paper improves on the basis of Faster R-CNN and uses the deep residual network Res2Net to enhance the original backbone network to improve the feature extraction ability. And the weighted feature fusion module is studied, which combines the high-level semantic information and the low-level high-resolution edge detail information to predict the feature map of each layer respectively, to improve the detection performance, especially for small targets. The experimental data show that this method can effectively improve the accuracy and efficiency of welding defect detection.

Driven by the knowledge economy and digitalization, the scale of book logistics continues to expand. However, the quality inspection process in this field currently uses generic target detection models and rarely considers defect characteristics. Therefore, this paper proposes the Merge-YOLO model to address the three prominent characteristics of book packaging defects: low contrast, small-sized defects, and irregular shapes. Three improvements are made to enhance detection performance: the WT-C3k2 module is designed to separate high- and low-frequency features using wavelet transforms, combined with multi-level convolutions and a bottleneck structure to enhance feature extraction capabilities for small objects and complex lighting conditions, while expanding the receptive field and reducing semantic detail loss; introducing the QA Transformer, which uses a learnable transformation matrix to generate adaptive quadrilateral windows, breaking through the limitations of traditional fixed windows and improving the ability to capture features of irregular defects; and adopting the DySample dynamic upscaler, which replaces nearest-neighbor interpolation by dynamically adjusting the scaling ratio through an adaptive scope factor, reducing computational overhead while preserving pixel-level details. Experiments show that the model achieves 95.8% precision, 93.6% recall, and 94.1%mAP@0.5 on the book packaging defect dataset, outperforming the baseline model YOLOv11 and traditional algorithms in all metrics. This provides an efficient and accurate detection model for quality control in book supply chain packaging.

In order to overcome the shortcomings of existing electrowetting display defect detection models in terms of computational complexity, structural complexity, detection speed, and detection accuracy, this article proposes an improved YOLOv7-based electrowetting display defect detection model. The model effectively optimizes the detection performance of display defects, especially small target defects, by integrating GhostNetV2 modules, Acmix attention mechanisms, and NGWD (Normalized Gaussian Wasserstein Distance) Loss. At the same time, it reduces the parameter size of the network model and improves the inference efficiency of the network. This article evaluates the performance of an improved model using a self-constructed electrowetting display defect dataset. The experimental results show that the proposed improved model achieves an average detection rate (mAP) of 89.5% and an average inference time of 35.9 ms. Compared to the original network, the number of parameters and computational costs are reduced by 19.2% and 64.3%, respectively. Compared with current state-of-the-art detection network models, the proposed EW-YOLOv7 exhibits superior performance in detecting electrowetting display defects. This model helps to solve the problem of defect detection in industrial production of electrowetting display and assists the research team in quickly identifying the causes and locations of defects.

Accurately and efficiently detecting steel surface defects is a critical step in steel manufacturing. However, the compromise between the detection speed and accuracy remains a major challenge, especially for steel surface defects with large variations in the scale. To address the issue, an improved you only look once (YOLO) based detection model is proposed through the reinforcement of its backbone and neck. Firstly, for the reduction of the redundant parameters and also the improvement of the characterization ability of the model, an effective channel residual structure is adopted to construct a channel residual convolution module and channel residual cross stage partial module as components of the backbone network, respectively. They realize the extraction of both the shallow feature and multi-scale feature simultaneously under a small number of convolutional parameters. Secondly, in the neck of YOLO, a fusion-distribution strategy is employed, which extracts and fuses multi-scale feature maps from the backbone network to provide global information, and then distributes global information into local features of different branches through an inject attention mechanism, thus enhancing the feature gap between different branches. Then, a model called CRFD-YOLO is derived for the steel surface defect detection and localization for the situations where both speed and accuracy are demanding. Finally, extensive experimental validations are conducted to evaluate the performance of CRFD-YOLO. The validation results indicate that CRFD-YOLO achieves a satisfactory detection performance with a mean average precision of 81.3% on the NEU-DET and 71.1% on the GC10-DET. Additionally, CRFD-YOLO achieves a speed of 161 frames per second, giving a great potential in real-time detection and localization tasks.

The essence of the difficulties for weld surface detection is that there is a lot of interference information during detection. This study aims to enhance the detection accuracy while keeping great deployment capabilities of a detection model for weld surface defects. To achieve this goal, an improved Yolo-graph convolution head (GCH) model is proposed based on the stable and fast Yolo-v5. The improvements primarily involve introducing a graph convolution network combined with a self-attention mechanism in the head part (i.e. GCH). This component focuses on improving the insufficient recognition capability of convolutional neural networks for similar defects in complex environments. Furthermore, to address the presence of potentially ambiguous samples in complex welding environments, the label assignment strategy of simOTA is implemented to optimize the anchor frame. Additionally, a streamlined structure, aiming to improve model detection speed while minimizing performance impact, has been designed to enhance the applicability of the model. The results demonstrate that the cooperation of GCH and simOTA significantly improves the detection performance while maintaining the inference speed. These strategies lead to a 2.5% increase in mAP@0.5 and reduce the missing detection rates of weld and 8 types of defects by 32.9% and 84.1% respectively, surpassing other weld surface detection models. Furthermore, the impressive applicability of the model is verified across four scaled versions of Yolo-v5. Based on the proposed strategies, the FPS increases by more than 30 frames in the fast s and n versions of Yolo-v5. These results demonstrate the great potential of the model for industrial applications.

Visual inspection of the appearance defects on industrial products has always been a research hotspot pursued by industry and academia. Due to the lack of samples in the industrial defect dataset and the serious class imbalance, deep learning technology cannot be directly applied to industrial defect visual inspection to meet the real application needs. Transfer learning is a good choice to deal with insufficient samples. However, cross-dataset bias is unavoidable during simple knowledge transfer. We noticed that the appearance defects of industrial products are similar, and most defects can be classified as stains or texture jumps, which provides a research basis for building a universal and adaptive industrial defect detection model. In this article, based on the idea of model-agnostic meta-learning (MAML), we propose an adaptive industrial defect detection model through learning from multiple known industrial defect datasets and then transfer it to the novel anomaly detection tasks. In addition, the Siamese network is used to extract differential features to minimize the influence of defect types on model generalization, and can also highlight defect features and improve model detection performance. At the same time, we add a coordinate attention mechanism to the model, which realizes the feature enhancement of the region of interest in terms of two coordinate dimensions. In the simulation experiments, we construct and publish a visual defect dataset of injection molded bottle cups, termed BC defects, which can complement existing industrial defect visual data benchmarks. Simulation results based on BC defects dataset and other public datasets have demonstrated the effectiveness of the proposed general visual detection model for industrial defects. The dataset and code are available at https://github.com/zhg-SZPT/MeDetection.

In the Unmanned Aerial Vehicle (UAV) transmission line inspection images, the detection of defective small-size object such as bolts on towers is important and challenging. Although using multi-scale features of deep neural network has improved the performance, it is still inadequate in mining fine-grained associations between multi-scale features and dealing with the high similarity between normal and defective bolts. Therefore, this paper proposes an improved defective bolt detection model MARF-CCN, based on Region of Interest (RoI) feature fusion and Cascaded Classification Network (CCN). First, a mixed attention RoI fusion network is built to adaptively compute fine-grained weights for features at different scales of feature pyramid network and enhances the difference between foreground and background. Second, cascaded classification network is designed to divide the original classification results into more easily identifiable categories based on morphological features, which are rectified via a secondary classifier to reduce false detection. Third, this paper defines atypical defects based on occurrence frequency and utilizes Focal Loss to address the resulting imbalanced classification loss. Experiments show that MARF-CCN improves the Average Precision (AP) of defective bolts by 14.33% to 84.40% compared with the commonly used models.

By assuming that the magnetic charge is evenly distributed in the defect groove, the linear magnetic dipole is integrated in the depth direction of the defect, and the numerical calculation is carried out by using MATLAB to establish a two-dimensional magnetic dipole theoretical analysis model of double-correlated defects, which is suitable for analyzing the influence of stress on magnetic signal in the magnetic memory detection of rectangular and V-shaped combined macroscopic crack defects. To explore the influence factors and change rules of magnetic memory signal. In order to verify the correctness of the model analysis results, tensile tests were carried out on Q235 steel plate with double associated defects with different morphological characteristics, and the Hp(y) value of the normal component of the leakage magnetic field on the surface of the member was measured. The results show that the model based on magnetic dipole theory can explain some experimental phenomena and rules in magnetic memory detection.