Search

In the production of strip steel, surface defect identification is crucial for improving product quality and ensuring smooth subsequent processes. Existing technologies face challenges such as low detection efficiency and susceptibility to environmental noise. This article employs an automated deep learning method without requiring consideration of complex environmental changes and proposes an improved RepVGG (ViT‐RepVGG) model for surface defect identification. The model is based on the RepVGG architecture, and the study explores the impact of incorporating the self‐attention mechanism of ViT under various addition strategies on model performance. A comparison is made between the optimized model and classic network models, as well as recently published models, in terms of identification performance. The research also examines the performance variations of the model under different hyperparameter settings and its identification performance for six types of defects. The results indicate that adding the ViT module to stage 3 of the A1‐type RepVGG, with a learning rate, optimizer, and activation function set to 0.0001, Adam, and Gelu, respectively, yields the optimal ViT‐RepVGG model performance. These findings demonstrate the feasibility of enhancing classification performance by incorporating the self‐attention mechanism into neural networks, providing an effective foundation for the online identification of strip steel surface defects. [ABSTRACT FROM AUTHOR]

Abstract The effective identification of fruit tree leaf disease is of great practical significance to reduce pesticide spraying, improve fruit yield and realize ecological agriculture. Computer vision technology can be effectively identifying and prevent plant diseases and insect pests. However, the lack of consideration of disease diversity and accuracy of existing detection models hinders their application and development in the field of plant pest detection. This paper proposes an efficient detection model of apple leaf disease spot through the improvement of the traditional Yolov5 detection network called A-Net. In order to significantly increase the A-Net's detection speed and accuracy, the A-Net model applies the loss function Wise-IoU, which includes the attention mechanism and the dynamic focusing mechanism, to the Yolov5 network model. The RepVGG module is then used to replace the original model's convolution module. The experimental results show that the improved model effectively suppresses the growth of some error weights. Compared with several object detection models, the improved A-Net model has a Mean Average Precision across IoU threshold 0.5 and an accuracy of 92.7%, which fully proves that the improved A-Net model has more advantages in detecting apple leaf diseases.

Deploying deep convolutional neural networks on agricultural devices with limited resources is challenging due to their large number of parameters. Existing lightweight networks can alleviate this problem but suffer from low performance. To this end, we propose a novel lightweight network named YOLOv8-RCAA (YOLOv8-RepVGG-CBAM-Anchorfree-ATSS), aiming to locate and detect tea leaf diseases with high accuracy and performance. Specifically, we employ RepVGG to replace CSPDarkNet63 to enhance feature extraction capability and inference efficiency. Then, we introduce CBAM attention to FPN and PAN in the neck layer to enhance the model perception of channel and spatial features. Additionally, an anchor-based detection head is replaced by an anchor-free head to further accelerate inference. Finally, we adopt the ATSS algorithm to adapt the allocating strategy of positive and negative samples during training to further enhance performance. Extensive experiments show that our model achieves precision, recall, F1 score, and mAP of 98.23%, 85.34%, 91.33%, and 98.14%, outperforming the traditional models by 4.22~6.61%, 2.89~4.65%, 3.48~5.52%, and 4.64~8.04%, respectively. Moreover, this model has a near-real-time inference speed, which provides technical support for deploying on agriculture devices. This study can reduce labor costs associated with the detection and prevention of tea leaf diseases. Additionally, it is expected to promote the integration of rapid disease detection into agricultural machinery in the future, thereby advancing the implementation of AI in agriculture. [ABSTRACT FROM AUTHOR]

The acoustic input network based on the Conformer encoder has the problem of insufficient extraction of FBank speech information and missing channel feature information. This paper proposed an end-to-end method based on RepVGG-SE-Conformer for speech recognition to solve these problems. Firstly, the proposed model used the multi-branch structure of RepVGG to enhance the speech information extraction capability, and using the structural re-parameterization fused the multibranch into a single branch to reduce the computational complexity and speed up the model inference. Then, based on the squeeze-and-excitation network, the channel attention mechanism made up for the missing channel feature information to improve speech recognition accuracy. Finally, the experimental results on the public dataset Aishell-1 show that the proposed method's character error rate is reduced by 10.67% compared with Conformer, and the advancement of the method is verified. In addition, the proposed RepVGG-SE acoustic input network has good generalization ability in the end-to-end scene, which can effectively improve the overall performance of speech recognition models based on Transformer variants. [ABSTRACT FROM AUTHOR]

Copyright of Cotton Textile Technology is the property of Cotton Textile Technology Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Image denoising is a fundamental problem in computer vision (CV). Vision Transformer (ViT) is an improvement in CV after convolutional neural networks (CNNs). Recently, it has been demonstrated that the RepVGG network of the structural re-parameterization performs well in image tasks. Experiments indicate that the ViT-based uniformer transformer network has successfully balanced local and global information. As image denoising tasks require more local and global information about the image, a novel image denoising model named structural Re-parameterization Uniformer Transformer-UNet (Rep-UUNet) is proposed in this paper. The model is structural and re-parameterized the structure of the Uniformer Transformer network and uses the UNet skip connections to reconstruct the output image. To complete image downstream tasks, the RepVGG model which utilizes the image local information is used. Indicators such as PSNR, SSIM and others are used to assess the performance of image denoising. Experimental results demonstrate that our Rep-UUNet network model outperforms other five models. [ABSTRACT FROM AUTHOR]

Carrier-free ultra-wideband sensors have high penetrability anti-jamming solid ability, which is not easily affected by the external environment, such as weather. Also, it has good performance in the complex jungle environment. In this paper, we propose a jungle vehicle identification system based on a carrier-free ultra-wideband sensor. Firstly, a composite jungle environment with the target vehicle is modeled. From this model, the simulation obtains time-domain echoes under the excitation of carrier-free ultra-wideband sensor signals in different orientations. Secondly, the time-domain signals are transformed into MTF images through the Markov transfer field to show the statistical characteristics of the time-domain echoes. At the same time, we propose an improved RepVGG network. The structure of the RepVGG network contains five stages, which consist of several RepVGG Blocks. Each RepVGG Block is created by combining convolutional kernels of different sizes using a weighted sum. We add the self-attention module to the output of stage 0 to improve the ability to extract the features of the MTF map and better capture the complex relationship between characteristics during training. In addition, a self-attention module is added before the linear layer classification output in stage 4 to improve the classification accuracy of the network. Moreover, a combined cross-entropy loss and sparsity penalty loss function helps enhance the performance and accuracy of the network. The experimental results show that the system can recognize jungle vehicle targets well. [ABSTRACT FROM AUTHOR]

In order to achieve the safety of autonomous driving, it is crucial to accurately detect traffic signs. Addressing the issue of low accuracy in traditional methods for traffic sign detection, a modified YOLOv5 algorithm incorporating the RepVGG module for traffic sign recognition was proposed. Firstly, by replacing certain CBS modules in the original algorithm with the RepVGG module, the feature extraction capability was enhanced. Additionally, the channel block attention module(CBAM) attention mechanism was introduced in the feature fusion module to strengthen the detection model's resistance to interference. Finally, during network training, the efficient-IoU(EIoU) loss function was utilized to achieve more precise target positioning, thereby improving the algorithm's detection accuracy and iteration speed. Experimental results demonstrate that the improved YOLOv5 algorithm exhibits faster iteration speed. On the CCTSDB traffic sign dataset, it achieves improvements of 4. 99% in accuracy, 3. 62% in recall rate, and 1. 73% in average accuracy compared to the original YOLOv5 algorithm, making it more suitable for practical applications. [ABSTRACT FROM AUTHOR]

In response to the safety concerns caused by falls and the limitations of existing object detection models in robustness, generalization and susceptibility to missing fall events, an optimized algorithm for detecting falls in the elderly population was proposed. The improved RepVGG modules was used for replacing the 3×3 convolutional modules of the YOLOv5s algorithm, the loss function was optimized, and the K-means++ clustering algorithm was employed to enhance dataset clustering. Experimental results demonstrate that the proposed algorithm exhibits strong robustness and generalization, achieving an average accuracy improvement of 9%, 8%, 3%, and 12% compared to YOLOv3, YOLOv4, YOLOv5s and CBAM-YOLOv5s models, respectively. The proposed algorithm meets the diverse requirements for fall detection in different real-world scenarios and can be applied in mobile devices or monitoring equipment, making a significant contribution to the field of elderly safety and protection.

Deploying deep convolutional neural networks on agricultural devices with limited resources is challenging due to their large number of parameters. Existing lightweight networks can alleviate this problem but suffer from low performance. To this end, we propose a novel lightweight network named YOLOv8-RCAA (YOLOv8-RepVGG-CBAM-Anchorfree-ATSS), aiming to locate and detect tea leaf diseases with high accuracy and performance. Specifically, we employ RepVGG to replace CSPDarkNet63 to enhance feature extraction capability and inference efficiency. Then, we introduce CBAM attention to FPN and PAN in the neck layer to enhance the model perception of channel and spatial features. Additionally, an anchor-based detection head is replaced by an anchor-free head to further accelerate inference. Finally, we adopt the ATSS algorithm to adapt the allocating strategy of positive and negative samples during training to further enhance performance. Extensive experiments show that our model achieves precision, recall, F1 score, and mAP of 98.23%, 85.34%, 91.33%, and 98.14%, outperforming the traditional models by 4.22~6.61%, 2.89~4.65%, 3.48~5.52%, and 4.64~8.04%, respectively. Moreover, this model has a near-real-time inference speed, which provides technical support for deploying on agriculture devices. This study can reduce labor costs associated with the detection and prevention of tea leaf diseases. Additionally, it is expected to promote the integration of rapid disease detection into agricultural machinery in the future, thereby advancing the implementation of AI in agriculture.

With the rapid development of Internet technology, the number of global Internet users is rapidly increasing, and the scale of the Internet is also expanding. The huge Internet system has accelerated the spread of bad information, including bad images. Bad images reflect the vulgar culture of the Internet. They will not only pollute the Internet environment and impact the core culture of society but also endanger the physical and mental health of young people. In addition, some criminals use bad images to induce users to download software containing computer viruses, which also greatly endanger the security of cyberspace. Cyberspace governance faces enormous challenges. Most existing methods for classifying bad images face problems such as low classification accuracy and long inference times, and these limitations are not conducive to effectively curbing the spread of bad images and reducing their harm. To address this issue, this paper proposes a classification method (RepVGG-SimAM) based on RepVGG and a simple parameter-free attention mechanism (SimAM). This method uses RepVGG as the backbone network and embeds the SimAM attention mechanism in the network so that the neural network can obtain more effective information and suppress useless information. We used pornographic images publicly disclosed by data scientist Alexander Kim and violent images collected from the internet to construct the dataset for our experiment. The experimental results prove that the classification accuracy of the method proposed in this paper can reach 94.5% for bad images, that the false positive rate of bad images is only 4.3%, and that the inference speed is doubled compared with the ResNet101 network. Our proposed method can effectively identify bad images and provide efficient and powerful support for cyberspace governance. [ABSTRACT FROM AUTHOR]

To further improve the detection of smoke and small target smoke in complex backgrounds, a novel smoke detection model called RepVGG-YOLOv7 is proposed in this paper. Firstly, the ECA attention mechanism and SIoU loss function are applied to the YOLOv7 network. The network effectively extracts the feature information of small targets and targets in complex backgrounds. Also, it makes the convergence of the loss function more stable and improves the regression accuracy. Secondly, RepVGG is added to the YOLOv7 backbone network to enhance the ability of the model to extract features in the training phase, while achieving lossless compression of the model in the inference phase. Finally, an improved non-maximal suppression algorithm is used to improve the detection in the case of dense smoke. Numerical experiments show that the detection accuracy of the proposed algorithm can reach about 95.1%, which contributes to smoke detection in complex backgrounds and small target smoke. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Hebei University of Science & Technology is the property of Hebei University of Science & Technology, Journal of Hebei University of Science & Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Transmission line fault detection using drones provides real-time assessment of the operational status of transmission equipment, and therefore it has immense importance in ensuring stable functioning of the transmission lines. Currently, identification of transmission line equipment relies predominantly on manual inspections that are susceptible to the influence of natural surroundings, resulting in sluggishness and a high rate of false detections. In view of this, in this study, we propose an insulator defect recognition algorithm based on a YOLOv5 model with a new lightweight network as the backbone network, combining noise reduction and target detection. First, we propose a new noise reduction algorithm, i.e., the adaptive neighborhood-weighted median filtering (NW-AMF) algorithm. This algorithm employs a weighted summation technique to determine the median value of the pixel point's neighborhood, effectively filtering out noise from the captured aerial images. Consequently, this approach significantly mitigates the adverse effects of varying noise levels on target detection. Subsequently, the RepVGG lightweight network structure is improved to the newly proposed lightweight structure called RcpVGG-YOLOv5. This structure facilitates single-branch inference, multi-branch training, and branch normalization, thereby improving the quantization performance while simultaneously striking a balance between target detection accuracy and speed. Furthermore, we propose a new loss function, i.e., Focal EIOU, to replace the original CIOU loss function. This optimization incorporates a penalty on the edge length of the target frame, which improves the contribution of the high-quality target gradient. This modification effectively addresses the issue of imbalanced positive and negative samples for small targets, suppresses background positive samples, and ultimately enhances the accuracy of detection. Finally, to align more closely with real-world engineering applications, the dataset utilized in this study consists of machine patrol images captured by the Unmanned Aerial Systems (UAS) of the Yunnan Power Supply Bureau Company. The experimental findings demonstrate that the proposed algorithm yields notable improvements in accuracy and inference speed compared to YOLOv5s, YOLOv7, and YOLOv8. Specifically, the improved algorithm achieves a 3.7% increase in accuracy and a 48.2% enhancement in inference speed compared to those of YOLOv5s. Similarly, it achieves a 2.7% accuracy improvement and a 33.5% increase in inference speed compared to those of YOLOv7, as well as a 1.5% accuracy enhancement and a 13.1% improvement in inference speed compared to those of YOLOv8. These results validate the effectiveness of the proposed algorithm through ablation experiments. Consequently, the method presented in this paper exhibits practical applicability in the detection of aerial images of transmission lines within complex environments. In future research endeavors, it is recommended to continue collecting aerial images for continuous iterative training, to optimize the model further, and to conduct in-depth investigations into the challenges associated with detecting small targets. Such endeavors hold significant importance for the advancement of transmission line detection. [ABSTRACT FROM AUTHOR]

Grape has been one of the most popular fruits with great nutritional value and economic benefits. Manual picking of mature grapes cannot fully meet the large-scale production in recent years, particularly with the expansion of planting areas. A picking robot can be expected to monitor the growth of grapes in orchards in real time. Automatic grape picking can also be promoted to realize intelligent agricultural production. In this study, an improved YOLOv5s model (MRW-YOLOv5s) was proposed to rapidly and accurately identify the grapes in orchards. Firstly, the lightweight network MobileNetv3 was used as the feature extraction network, in order to reduce the amount of model parameters. A coordinate attention module (CA) was also embedded into the bneck structure of MobileNetv3 to strengthen the feature extraction capability of the network. Secondly, RepVGG Block was introduced into the neck network, where the multi-branch features were integrated to improve the detection accuracy of the model. Moreover, the structural reparameterization of the RepVGG Block was implemented to further accelerate the inference speed of the model. Finally, Wise Intersection over Union Loss (WIoU Loss) with the dynamic nonmonotonic focusing mechanism was taken as the bounding box regression loss function, in order to accelerate the network convergence for the better detection accuracy of the model. Gradient-weighted class activation mapping (Grad-CAM) was also selected to capture the grape targets when the backbone network of the improved model was embedded with the CA module. A better performance was then achieved, compared with the model embedded with Efficient Channel Attention (ECA) and Convolutional Block Attention Module (CBAM). In addition, there was the lowest speed of bounding box loss regression in the convergence curve of the loss function, while the highest loss value after convergence, where the EIoU was the bounding box loss function. Once the CIoU and Wise-IoU v1 were taken as the bounding box loss functions, there were similar convergence speeds and loss values, indicating a slightly lower value than that of EIoU. Moreover, there was the highest convergence speed of the model and the lowest loss value, when the Wise-IoU v3 was used as the bounding box loss function. Therefore, the Wise-IoU v3 can be expected to accelerate the convergence of the model for better accuracy of the model. The results showed that the number of parameters of the improved MRW-YOLOv5s model was only 7.56 M. The mean Average Precision (mAP) on the test set reached 97.74%, and the average detection time per image was 10.03 ms, which were 2.32 percentage points and 6ms higher than those of the original YOLOv5s model, respectively. The mAP of the MRW-YOLOv5s model was 9.89, 7.53, 2.12, 0.91, and 2.42 percentage points higher, respectively, compared with the mainstream object detection models, such as the SSD, RetinaNet, YOLOv4, YOLOv7, and YOLOX. In terms of the number of model parameters, the improved model was only 7.56 M, which was 68.2%, 79.2%, 88.2%, 79.7%, and 15.4% less than the above five models, respectively. The average detection speed of the improved model was only 10.03 ms, which was 2.64, 13.19, 10.59, 4.14, and 5.46 ms higher than the above five models, respectively. Furthermore, the weight size of the improved model was only 26.97 MB, which was more conducive to model deployment. Therefore, the MRW-YOLOv5s model can greatly contribute to the detection accuracy, parameter size, and detection speed. The finding can also provide technical support for the intelligent orchards and mechanization of picking. [ABSTRACT FROM AUTHOR]

Carrier-free ultra-wideband sensors have high penetrability anti-jamming solid ability, which is not easily affected by the external environment, such as weather. Also, it has good performance in the complex jungle environment. In this paper, we propose a jungle vehicle identification system based on a carrier-free ultra-wideband sensor. Firstly, a composite jungle environment with the target vehicle is modeled. From this model, the simulation obtains time-domain echoes under the excitation of carrier-free ultra-wideband sensor signals in different orientations. Secondly, the time-domain signals are transformed into MTF images through the Markov transfer field to show the statistical characteristics of the time-domain echoes. At the same time, we propose an improved RepVGG network. The structure of the RepVGG network contains five stages, which consist of several RepVGG Blocks. Each RepVGG Block is created by combining convolutional kernels of different sizes using a weighted sum. We add the self-attention module to the output of stage 0 to improve the ability to extract the features of the MTF map and better capture the complex relationship between characteristics during training. In addition, a self-attention module is added before the linear layer classification output in stage 4 to improve the classification accuracy of the network. Moreover, a combined cross-entropy loss and sparsity penalty loss function helps enhance the performance and accuracy of the network. The experimental results show that the system can recognize jungle vehicle targets well.

IntroductionPneumonia is a common and widespread infectious disease that seriously affects the life and health of patients. Especially in recent years, the outbreak of COVID-19 has caused a sharp rise in the number of confirmed cases of epidemic spread. Therefore, early detection and treatment of pneumonia are very important. However, the uneven gray distribution and structural intricacy of pneumonia images substantially impair the classification accuracy of pneumonia. In this classification task of COVID-19 and other pneumonia, because there are some commonalities between this pneumonia, even a small gap will lead to the risk of prediction deviation, it is difficult to achieve high classification accuracy by directly using the current network model to optimize the classification model.MethodsConsequently, an optimization method for the CT classification model of COVID-19 based on RepVGG was proposed. In detail, it is made up of two essential modules, feature extraction backbone and spatial attention block, which allows it to extract spatial attention features while retaining the benefits of RepVGG.ResultsThe model’s inference time is significantly reduced, and it shows better learning ability than RepVGG on both the training and validation sets. Compared with the existing advanced network models VGG-16, ResNet-50, GoogleNet, ViT, AlexNet, MobileViT, ConvNeXt, ShuffleNet, and RepVGG_b0, our model has demonstrated the best performance in a lot of indicators. In testing, it achieved an accuracy of 0.951, an F1 score of 0.952, and a Youden index of 0.902.DiscussionOverall, multiple experiments on the large dataset of SARS-CoV-2 CT-scan dataset reveal that this method outperforms most basic models in terms of classification and screening of COVID-19 CT, and has a significant reference value. Simultaneously, in the inspection experiment, this method outperformed other networks with residual structures.

Simple Summary: Identifying individual sheep accurately is crucial for establishing precise animal husbandry. In the process of identifying sheep by their faces, changes in sheep face poses and different camera angles can affect the identification accuracy. In this study, we construct a new sheep face recognition model. Sheep face data with different poses and angles are used as input in a bilinear feature extraction network, which extracts the important features of sheep faces separately. Then, a feature fusion method is used to fuse the features extracted by the bilinear network for sheep face recognition. Our experimental results demonstrate that the recognition accuracy of the algorithm is 99.43%, achieving the individual recognition of sheep in complex environments while reducing the influence of pose and angle on recognition. A key prerequisite for the establishment of digitalized sheep farms and precision animal husbandry is the accurate identification of each sheep's identity. Due to the uncertainty in recognizing sheep faces, the differences in sheep posture and shooting angle in the recognition process have an impact on the recognition accuracy. In this study, we propose a deep learning model based on the RepVGG algorithm and bilinear feature extraction and fusion for the recognition of sheep faces. The model training and testing datasets consist of photos of sheep faces at different distances and angles. We first design a feature extraction channel with an attention mechanism and RepVGG blocks. The RepVGG block reparameterization mechanism is used to achieve lossless compression of the model, thus improving its recognition efficiency. Second, two feature extraction channels are used to form a bilinear feature extraction network, which extracts important features for different poses and angles of the sheep face. Finally, features at the same scale from different images are fused to enhance the feature information, improving the recognition ability and robustness of the network. The test results demonstrate that the proposed model can effectively reduce the effect of sheep face pose on the recognition accuracy, with recognition rates reaching 95.95%, 97.64%, and 99.43% for the sheep side-, front-, and full-face datasets, respectively, outperforming several state-of-the-art sheep face recognition models. [ABSTRACT FROM AUTHOR]

With the continuous advancement of remote sensing technology, satellite remote sensing images have become one of the important means of obtaining information on the earth surface. But in the current research on aircraft target detection and classification in remote sensing images, the imbalanced data samples, large variations in target scales and backgrounds, and target occlusion have led to low average precision and slow detection speed in detection and classification tasks. Therefore, this paper proposes the YOLO-class model. Firstly, the YOLO-Extract model is transferred to optimize the detection of small targets, dense targets, and occluded targets. Secondly, Representative Batch Normalization and Mish activation function are used to optimize the Conv module, and VariFocal loss is used to optimize the classification loss function to improve the accuracy caused by imbalanced data samples. Finally, RepVGG modules are designed in the Backbone to further improve the detection accuracy of the model. Simulation results show that compared with the YOLO-Extract model, YOLO-class improves the detection accuracy from 0.608 to 0.704 and FPS from 36.16 to 39.598.

Wind turbine blades will suffer various surface damages due to their operating environment and high-speed rotation. Accurate identification in the early stage of damage formation is crucial. The damage detection of wind turbine blades is a primarily manual operation, which has problems such as high cost, low efficiency, intense subjectivity, and high risk. The rise of deep learning provides a new method for detecting wind turbine blade damage. However, in detecting wind turbine blade damage in general network models, there will be an insufficient fusion of multiscale small target features. This paper proposes a lightweight cascaded feature fusion neural network model based on YOLOX. Firstly, the lightweight area of the backbone feature extraction network concerning the RepVGG network structure is enhanced, improving the model's inference speed. Second, a cascaded feature fusion module is designed to cascade and interactively fuse multilevel features to enhance the small target area features and the model's feature perception capabilities for multiscale target damage. The focal loss is introduced in the post-processing stage to enhance the network's ability to learn complex positive sample damages. The detection accuracy of the improved algorithm is increased by 2.95%, the mAP can reach 94.29% in the self-made dataset, and the recall rate and detection speed are slightly improved. The experimental results show that the algorithm can autonomously learn the blade damage features from the wind turbine blade images collected in the actual scene, achieve the automatic detection, location, and classification of wind turbine blade damage, and promote the detection of wind turbine blade damage towards automation, rapidity, and low-cost development. [ABSTRACT FROM AUTHOR]

With the worldwide carbon neutralization boom, low-speed heavy load bearings have been widely used in the field of wind power. Bearing failure generates impulses when the rolling element passes the cracked surface of the bearing. Over the past decade, acoustic emission (AE) techniques have been used to detect failure signals. However, the high sampling rates of AE signals make it difficult to design and extract fault features; thus, deep neural network-based approaches have been proposed. In this paper, we proposed an improved RepVGG bearing fault diagnosis technique. The normalized and noise-reduced bearing signals were first converted into Mel frequency cepstrum coefficients (MFCCs) and then inputted into the model. In addition, the exponential moving average method was used to optimize the model and improve its accuracy. Data were extracted from the test bench and wind turbine main shaft bearing. Four damage classes were studied experimentally. The experimental results demonstrated that the improved RepVGG model could be employed for classifying low-speed heavy load bearing states by using MFCCs. Furthermore, the effectiveness of the proposed model was assessed by performing comparisons with existing models. [ABSTRACT FROM AUTHOR]

Nowadays, the detection model based on YOLOv5 network are widely used in the task of pedestrian detection, and have good results in terms of accuracy and speed. However, it is often limited by computing power on terminal devices. Therefore, based on the improved backbone network of the RepVGG model, in order to improve the adaptability in dense crowds and complex environments, a coordinate attention mechanism was introduced to expand the receptive field and enhance the weight of the region of interest. After experimental testing, this lightweight network parameter and calculation amount is relatively small, and the detection accuracy and robustness are relatively high, which can meet the requirements of engineering applications to a certain extent. [ABSTRACT FROM AUTHOR]

Underwater object detection currently faces many challenges, such as the large number of parameters in existing object detection models, slow inference speed, blurring of underwater images, and aggregation of small targets, making it difficult to conduct efficient underwater object detection. This paper proposes a lightweight underwater object detection algorithm based on YOLOv5.The method uses depth-wise separable convolution instead of ordinary convolution to reduce the number of parameters and computational complexity. A C3 module based on Ghost convolution is designed to further compress the model size and improve the computational speed. In the feature extraction stage, a RepVgg module based on structural reparameterization is used to convert the multi -branch structure into a single-branch structure in the inference stage, improving the feature extraction ability of the model and increasing the inference speed. A Rep-ECA module is designed to embed the efficient channel attention module ECANet into the RepVGG module, selecting more effective channel information and improving the model’s feature extraction ability for small objects in blurred images, thereby improving detection precision. Experimental results show that in the URPC underwater object detection dataset, the proposed algorithm has a 39% lower model parameter count compared to the original model, a 42% reduction in computational complexity. The model can achieve a frame rate of 85 on a single Nvidia GTX 1080ti GPU, which is a 24% improvement over the original model, while mAP reaches 85.1%, a 1.1% improvement over the original model. The algorithm can improve the detection precision and achieve lightweight, which lays a foundation for the deployment of underwater equipment.

Focus on the requirement for detecting laser welding defects of lithium battery pole, a new model based on the improved YOLOv5 algorithm was proposed in this paper. First, all the 3 × 3 convolutional kernels in the backbone network were replaced by 6 × 6 convolutional kernels to improve the model's detection capability of a small defect; second, the last layer of the backbone network was replaced by our designed SPPSE module to enhance the detection accuracy of the model; then the improved RepVGG module was introduced in the head network, which can help to improve the inference speed of the model and enhance the feature extraction capability of the network; finally, SIOU was used as the bounding box regression loss function to improve the accuracy and training speed of the model. The experimental results show that our improved YOLOv5 model achieved 97% mAP and 270 fps on our dataset. Compared with conventional methods, ours had the best results. The ablation experiments were conducted on the publicly available datasets PASCAL VOC and MS COCO, and their mAP@0.5 was improved by 2.4% and 3%, respectively. Additionally, our model improved the average detection rate for small targets on the MS COCO dataset by 2.4%, showing that it can effectively detect small target defects. [ABSTRACT FROM AUTHOR]

The attitude sensor of the aircraft can give feedback on the perceived flight attitude information to the input of the flight controller to realize the closed-loop control of the flight attitude. Therefore, the fault diagnosis of attitude sensors is crucial for the flight safety of aircraft, in view of the situation that the existing diagnosis methods fail to give consideration to both the diagnosis rate and the diagnosis accuracy. In this paper, a fast and high-precision fault diagnosis strategy for aircraft sensor is proposed. Specifically, the aircraft's dynamics model and the attitude sensor's fault model are built. The SENet attention mechanism is used to allocate weights for the collected time-domain fault signals and transformed time-frequency signals, and then inject the fused feature signals with weights into the RepVGG based on the convolutional neural network structure for deep feature mining and classification. Experimental results show that the proposed method can achieve good precision speed tradeoff. [ABSTRACT FROM AUTHOR]

With the rapid development of Internet technology, the number of global Internet users is rapidly increasing, and the scale of the Internet is also expanding. The huge Internet system has accelerated the spread of bad information, including bad images. Bad images reflect the vulgar culture of the Internet. They will not only pollute the Internet environment and impact the core culture of society but also endanger the physical and mental health of young people. In addition, some criminals use bad images to induce users to download software containing computer viruses, which also greatly endanger the security of cyberspace. Cyberspace governance faces enormous challenges. Most existing methods for classifying bad images face problems such as low classification accuracy and long inference times, and these limitations are not conducive to effectively curbing the spread of bad images and reducing their harm. To address this issue, this paper proposes a classification method (RepVGG-SimAM) based on RepVGG and a simple parameter-free attention mechanism (SimAM). This method uses RepVGG as the backbone network and embeds the SimAM attention mechanism in the network so that the neural network can obtain more effective information and suppress useless information. We used pornographic images publicly disclosed by data scientist Alexander Kim and violent images collected from the internet to construct the dataset for our experiment. The experimental results prove that the classification accuracy of the method proposed in this paper can reach 94.5% for bad images, that the false positive rate of bad images is only 4.3%, and that the inference speed is doubled compared with the ResNet101 network. Our proposed method can effectively identify bad images and provide efficient and powerful support for cyberspace governance.

To further improve the detection of smoke and small target smoke in complex backgrounds, a novel smoke detection model called RepVGG-YOLOv7 is proposed in this paper. Firstly, the ECA attention mechanism and SIoU loss function are applied to the YOLOv7 network. The network effectively extracts the feature information of small targets and targets in complex backgrounds. Also, it makes the convergence of the loss function more stable and improves the regression accuracy. Secondly, RepVGG is added to the YOLOv7 backbone network to enhance the ability of the model to extract features in the training phase, while achieving lossless compression of the model in the inference phase. Finally, an improved non-maximal suppression algorithm is used to improve the detection in the case of dense smoke. Numerical experiments show that the detection accuracy of the proposed algorithm can reach about 95.1%, which contributes to smoke detection in complex backgrounds and small target smoke.

Transmission line fault detection using drones provides real-time assessment of the operational status of transmission equipment, and therefore it has immense importance in ensuring stable functioning of the transmission lines. Currently, identification of transmission line equipment relies predominantly on manual inspections that are susceptible to the influence of natural surroundings, resulting in sluggishness and a high rate of false detections. In view of this, in this study, we propose an insulator defect recognition algorithm based on a YOLOv5 model with a new lightweight network as the backbone network, combining noise reduction and target detection. First, we propose a new noise reduction algorithm, i.e., the adaptive neighborhood-weighted median filtering (NW-AMF) algorithm. This algorithm employs a weighted summation technique to determine the median value of the pixel point’s neighborhood, effectively filtering out noise from the captured aerial images. Consequently, this approach significantly mitigates the adverse effects of varying noise levels on target detection. Subsequently, the RepVGG lightweight network structure is improved to the newly proposed lightweight structure called RcpVGG-YOLOv5. This structure facilitates single-branch inference, multi-branch training, and branch normalization, thereby improving the quantization performance while simultaneously striking a balance between target detection accuracy and speed. Furthermore, we propose a new loss function, i.e., Focal EIOU, to replace the original CIOU loss function. This optimization incorporates a penalty on the edge length of the target frame, which improves the contribution of the high-quality target gradient. This modification effectively addresses the issue of imbalanced positive and negative samples for small targets, suppresses background positive samples, and ultimately enhances the accuracy of detection. Finally, to align more closely with real-world engineering applications, the dataset utilized in this study consists of machine patrol images captured by the Unmanned Aerial Systems (UAS) of the Yunnan Power Supply Bureau Company. The experimental findings demonstrate that the proposed algorithm yields notable improvements in accuracy and inference speed compared to YOLOv5s, YOLOv7, and YOLOv8. Specifically, the improved algorithm achieves a 3.7% increase in accuracy and a 48.2% enhancement in inference speed compared to those of YOLOv5s. Similarly, it achieves a 2.7% accuracy improvement and a 33.5% increase in inference speed compared to those of YOLOv7, as well as a 1.5% accuracy enhancement and a 13.1% improvement in inference speed compared to those of YOLOv8. These results validate the effectiveness of the proposed algorithm through ablation experiments. Consequently, the method presented in this paper exhibits practical applicability in the detection of aerial images of transmission lines within complex environments. In future research endeavors, it is recommended to continue collecting aerial images for continuous iterative training, to optimize the model further, and to conduct in-depth investigations into the challenges associated with detecting small targets. Such endeavors hold significant importance for the advancement of transmission line detection.

A key prerequisite for the establishment of digitalized sheep farms and precision animal husbandry is the accurate identification of each sheep’s identity. Due to the uncertainty in recognizing sheep faces, the differences in sheep posture and shooting angle in the recognition process have an impact on the recognition accuracy. In this study, we propose a deep learning model based on the RepVGG algorithm and bilinear feature extraction and fusion for the recognition of sheep faces. The model training and testing datasets consist of photos of sheep faces at different distances and angles. We first design a feature extraction channel with an attention mechanism and RepVGG blocks. The RepVGG block reparameterization mechanism is used to achieve lossless compression of the model, thus improving its recognition efficiency. Second, two feature extraction channels are used to form a bilinear feature extraction network, which extracts important features for different poses and angles of the sheep face. Finally, features at the same scale from different images are fused to enhance the feature information, improving the recognition ability and robustness of the network. The test results demonstrate that the proposed model can effectively reduce the effect of sheep face pose on the recognition accuracy, with recognition rates reaching 95.95%, 97.64%, and 99.43% for the sheep side-, front-, and full-face datasets, respectively, outperforming several state-of-the-art sheep face recognition models.

As one of the fatal human diseases, early detection of brain tumors can effectively save patients' lives. Brain tumor image segmentation is of great practical importance for physicians to perform brain tumor diagnoses quickly. Due to the data complexity of 3D brain images, it is impractical to segment out tumor regions manually, so automatic and reliable methods can be utilized instead of manual work to achieve accurate segmentation of tumor regions. In this paper, we propose an end-to-end, more efficient brain tumor MRI segmentation model, REMU-Net, for the problems of multi-scale feature extraction and difficulty in small target feature extraction in 3D brain tumor image segmentation. Firstly, design and use the multi-channel parallel M-RepVGG module as a decoder to achieve multi-scale feature fusion. Secondly, embedding dilated convolution with different dilated rates in the DM-RepVGG module of the encoder to better extract features at different scales. Finally, introduce the expectation-maximizing attention in the network to better extract the features of the internal details of the tumor. The experimental results on the BraTS2018 validation dataset are Dice scores of 80.93%, 90.13%, and 86.15%, respectively. Experimental results on the BraTS2019 validation dataset can be achieved with Dice scores of 78.29%, 90.65%, and 82.77%, respectively. [ABSTRACT FROM AUTHOR]

Oil storage tank detection and classification in synthetic aperture radar (SAR) images play a vital role in monitoring energy distribution and consumption. Due to the SAR side-looking imaging geometry and multibouncing scattering mechanism, dense oil tank detection and classification tasks have faced more challenges, such as overlapping, blurred contours, and geometric distortion, especially for small-sized tanks. To address the above issues, this paper proposes YOLOX-TR, an improved YOLOX based on the Transformer encoder and structural reparameterized VGG-like (RepVGG) blocks, to achieve end-to-end oil tank detection and classification in densely arranged areas of large-scale SAR images. Based on YOLOX, the Transformer encoder, a self-attention-based architecture, is integrated to enhance the representation of feature maps and capture the region of interest of oil tanks in densely distributed scenarios. Furthermore, RepVGG blocks are employed to reparameterize the backbone with multibranch typologies to strengthen the distinguishable feature extraction of multi-scale oil tanks without increasing computation in inference time. Eventually, comprehensive experiments based on a Gaofen-3 1 m oil tank dataset (OTD) demonstrated the effectiveness of the Transformer encoder and RepVGG blocks, as well as the performance superiority of YOLOX-TR with a mAP and mAP0.5 of 60.8% and 94.8%, respectively. [ABSTRACT FROM AUTHOR]

Wind turbine blades will suffer various surface damages due to their operating environment and high-speed rotation. Accurate identification in the early stage of damage formation is crucial. The damage detection of wind turbine blades is a primarily manual operation, which has problems such as high cost, low efficiency, intense subjectivity, and high risk. The rise of deep learning provides a new method for detecting wind turbine blade damage. However, in detecting wind turbine blade damage in general network models, there will be an insufficient fusion of multiscale small target features. This paper proposes a lightweight cascaded feature fusion neural network model based on YOLOX. Firstly, the lightweight area of the backbone feature extraction network concerning the RepVGG network structure is enhanced, improving the model’s inference speed. Second, a cascaded feature fusion module is designed to cascade and interactively fuse multilevel features to enhance the small target area features and the model’s feature perception capabilities for multiscale target damage. The focal loss is introduced in the post-processing stage to enhance the network’s ability to learn complex positive sample damages. The detection accuracy of the improved algorithm is increased by 2.95%, the mAP can reach 94.29% in the self-made dataset, and the recall rate and detection speed are slightly improved. The experimental results show that the algorithm can autonomously learn the blade damage features from the wind turbine blade images collected in the actual scene, achieve the automatic detection, location, and classification of wind turbine blade damage, and promote the detection of wind turbine blade damage towards automation, rapidity, and low-cost development.

With the worldwide carbon neutralization boom, low-speed heavy load bearings have been widely used in the field of wind power. Bearing failure generates impulses when the rolling element passes the cracked surface of the bearing. Over the past decade, acoustic emission (AE) techniques have been used to detect failure signals. However, the high sampling rates of AE signals make it difficult to design and extract fault features; thus, deep neural network-based approaches have been proposed. In this paper, we proposed an improved RepVGG bearing fault diagnosis technique. The normalized and noise-reduced bearing signals were first converted into Mel frequency cepstrum coefficients (MFCCs) and then inputted into the model. In addition, the exponential moving average method was used to optimize the model and improve its accuracy. Data were extracted from the test bench and wind turbine main shaft bearing. Four damage classes were studied experimentally. The experimental results demonstrated that the improved RepVGG model could be employed for classifying low-speed heavy load bearing states by using MFCCs. Furthermore, the effectiveness of the proposed model was assessed by performing comparisons with existing models.

Focus on the requirement for detecting laser welding defects of lithium battery pole, a new model based on the improved YOLOv5 algorithm was proposed in this paper. First, all the 3 × 3 convolutional kernels in the backbone network were replaced by 6 × 6 convolutional kernels to improve the model’s detection capability of a small defect; second, the last layer of the backbone network was replaced by our designed SPPSE module to enhance the detection accuracy of the model; then the improved RepVGG module was introduced in the head network, which can help to improve the inference speed of the model and enhance the feature extraction capability of the network; finally, SIOU was used as the bounding box regression loss function to improve the accuracy and training speed of the model. The experimental results show that our improved YOLOv5 model achieved 97% mAP and 270 fps on our dataset. Compared with conventional methods, ours had the best results. The ablation experiments were conducted on the publicly available datasets PASCAL VOC and MS COCO, and their mAP@0.5 was improved by 2.4% and 3%, respectively. Additionally, our model improved the average detection rate for small targets on the MS COCO dataset by 2.4%, showing that it can effectively detect small target defects.

The attitude sensor of the aircraft can give feedback on the perceived flight attitude information to the input of the flight controller to realize the closed-loop control of the flight attitude. Therefore, the fault diagnosis of attitude sensors is crucial for the flight safety of aircraft, in view of the situation that the existing diagnosis methods fail to give consideration to both the diagnosis rate and the diagnosis accuracy. In this paper, a fast and high-precision fault diagnosis strategy for aircraft sensor is proposed. Specifically, the aircraft’s dynamics model and the attitude sensor’s fault model are built. The SENet attention mechanism is used to allocate weights for the collected time-domain fault signals and transformed time-frequency signals, and then inject the fused feature signals with weights into the RepVGG based on the convolutional neural network structure for deep feature mining and classification. Experimental results show that the proposed method can achieve good precision speed tradeoff.

The automatic classification of aluminum profile surface defects is of great significance in improving the surface quality of aluminum profiles in practical production. This classification is influenced by the small and unbalanced number of samples and lack of uniformity in the size and spatial distribution of aluminum profile surface defects. It is difficult to achieve high classification accuracy by directly using the current advanced classification algorithms. In this paper, digital image processing methods such as rotation, flipping, contrast, and luminance transformation were used to augment the number of samples and imitate the complex imaging environment in actual practice. A RepVGG with CBAM attention mechanism (RepVGG-CBAM) model was proposed and applied to classify ten types of aluminum profile surface defects. The classification accuracy reached 99.41%, in particular, the proposed method can perfectly classify six types of defects: concave line (cl), exposed bottom (eb), exposed corner bottom (ecb), mixed color (mc), non-conductivity (nc) and orange peel (op), with 100% precision, recall, and F1. Compared with the existing advanced classification algorithms VGG16, VGG19, ResNet34, ResNet50, ShuffleNet_v2, and basic RepVGG, our model is the best in terms of accuracy, macro precision, macro recall and macro F1, and the accuracy was improved by 4.85% over basic RepVGG. Finally, an ablation experiment proved that the classification ability was strongest when the CBAM attention mechanism was added following Stage 1 to Stage 4 of RepVGG. Overall, the method we proposed in this paper has a significant reference value for classifying aluminum profile surface defects.

Oil storage tank detection and classification in synthetic aperture radar (SAR) images play a vital role in monitoring energy distribution and consumption. Due to the SAR side-looking imaging geometry and multibouncing scattering mechanism, dense oil tank detection and classification tasks have faced more challenges, such as overlapping, blurred contours, and geometric distortion, especially for small-sized tanks. To address the above issues, this paper proposes YOLOX-TR, an improved YOLOX based on the Transformer encoder and structural reparameterized VGG-like (RepVGG) blocks, to achieve end-to-end oil tank detection and classification in densely arranged areas of large-scale SAR images. Based on YOLOX, the Transformer encoder, a self-attention-based architecture, is integrated to enhance the representation of feature maps and capture the region of interest of oil tanks in densely distributed scenarios. Furthermore, RepVGG blocks are employed to reparameterize the backbone with multibranch typologies to strengthen the distinguishable feature extraction of multi-scale oil tanks without increasing computation in inference time. Eventually, comprehensive experiments based on a Gaofen-3 1 m oil tank dataset (OTD) demonstrated the effectiveness of the Transformer encoder and RepVGG blocks, as well as the performance superiority of YOLOX-TR with a mAP and mAP0.5 of 60.8% and 94.8%, respectively.

In the intelligent perception of the marine engine room, visual identification of auxiliary equipment is the prerequisite for defect recognition and anomaly detection. To improve the detection accuracy, this study presents an auxiliary equipment detector in the cabin based on a deep learning model. Owing to the compact layout of pipeline networks and the large disparity in the equipment scales, we initially adopted RetinaNet as the basic framework, and introduced the single channel plain architecture RepVGG as the feature extraction network to simplify the complexity and improve realtime detection. Secondly, the Neighbor Erasing and Transferring Mechanism (NETM) was applied in the feature pyramid to deal with more complicated scale variations. Then, the complete IoU (CIoU) regression loss function was used instead of smooth L1, and the DIoU Soft-NMS mechanism was proposed to alleviate the misdetection in congested cabins. Further, comparison experiments and ablation experiments were performed on the auxiliary equipment in a marine engine room (AEMER) dataset to validate the efficacy of these strategies on the model performance boost. Specifically, our model can correctly detect 93.44% of coolers, 100.00% of diesel engines, 60.26% of meters, 95.30% of pumps, 55.01% of reservoirs, 97.68% of oil separators, and 74.37% of valves in a practical cabin.

It is important to accurately classify the defects in hot rolled steel strip since the detection of defects in hot rolled steel strip is closely related to the quality of the final product. The lack of actual hot-rolled strip defect data sets currently limits further research on the classification of hot-rolled strip defects to some extent. In real production, the convolutional neural network (CNN)-based algorithm has some difficulties, for example, the algorithm is not particularly accurate in classifying some uncommon defects. Therefore, further research is needed on how to apply deep learning to the actual detection of defects on the surface of hot rolled steel strip. In this paper, we proposed a hot rolled steel strip defect dataset called Xsteel surface defect dataset (X-SDD) which contains seven typical types of hot rolled strip defects with a total of 1360 defect images. Compared with the six defect types of the commonly used NEU surface defect database (NEU-CLS), our proposed X-SDD contains more types. Then, we adopt the newly proposed RepVGG algorithm and combine it with the spatial attention (SA) mechanism to verify the effect on the X-SDD. Finally, we apply multiple algorithms to test on our proposed X-SDD to provide the corresponding benchmarks. The test results show that our algorithm achieves an accuracy of 95.10% on the testset, which exceeds other comparable algorithms by a large margin. Meanwhile, our algorithm achieves the best results in Macro-Precision, Macro-Recall and Macro-F1-score metrics.

Introduction: Pneumonia is a common and widespread infectious disease that seriously affects the life and health of patients. Especially in recent years, the outbreak of COVID-19 has caused a sharp rise in the number of confirmed cases of epidemic spread. Therefore, early detection and treatment of pneumonia are very important. However, the uneven gray distribution and structural intricacy of pneumonia images substantially impair the classification accuracy of pneumonia. In this classification task of COVID-19 and other pneumonia, because there are some commonalities between this pneumonia, even a small gap will lead to the risk of prediction deviation, it is difficult to achieve high classification accuracy by directly using the current network model to optimize the classification model., Methods: Consequently, an optimization method for the CT classification model of COVID-19 based on RepVGG was proposed. In detail, it is made up of two essential modules, feature extraction backbone and spatial attention block, which allows it to extract spatial attention features while retaining the benefits of RepVGG., Results: The model's inference time is significantly reduced, and it shows better learning ability than RepVGG on both the training and validation sets. Compared with the existing advanced network models VGG-16, ResNet-50, GoogleNet, ViT, AlexNet, MobileViT, ConvNeXt, ShuffleNet, and RepVGG_b0, our model has demonstrated the best performance in a lot of indicators. In testing, it achieved an accuracy of 0.951, an F1 score of 0.952, and a Youden index of 0.902., Discussion: Overall, multiple experiments on the large dataset of SARS-CoV-2 CT-scan dataset reveal that this method outperforms most basic models in terms of classification and screening of COVID-19 CT, and has a significant reference value. Simultaneously, in the inspection experiment, this method outperformed other networks with residual structures., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest., (Copyright © 2023 Zhang, Shu, Chen, Teng, Gu, Li and Kan.)

• A modified RepVGG based visual servo method for quick-change tool alignment is investigated. • The robot is controlled by the position-based visual servo control law. • Experiments in real world are conducted to validate the effectiveness of the proposed visual servo method. The remote maintenance process for fusion reactor is complex, which can be very time-consuming and labor-consuming. This paper proposes a modified neural network based visual servo method to align a quick-change device with robot camera system. A classical position-based visual servo control law is used to guide the robot to reach the desired position. The key target for the above visual servo controller is to obtain a robust pose estimator to calculate the quick-change device pose with respect to the camera. This pose estimator is trained using RepVGG model with a self built image samples. An attention mechanism is added to the neural network to enhance the stability of pose prediction for reflective metal objects. The robot joint speed is also smoothed to reduce the image motion blur effect and make the visual servo process stable. The performance of the proposed visual servo controller was verified on an UR5 robot, and the results show that the stable and rough alignment of the quick-change device can be realized. [ABSTRACT FROM AUTHOR]

In the intelligent perception of the marine engine room, visual identification of auxiliary equipment is the prerequisite for defect recognition and anomaly detection. To improve the detection accuracy, this study presents an auxiliary equipment detector in the cabin based on a deep learning model. Owing to the compact layout of pipeline networks and the large disparity in the equipment scales, we initially adopted RetinaNet as the basic framework, and introduced the single channel plain architecture RepVGG as the feature extraction network to simplify the complexity and improve realtime detection. Secondly, the Neighbor Erasing and Transferring Mechanism (NETM) was applied in the feature pyramid to deal with more complicated scale variations. Then, the complete IoU (CIoU) regression loss function was used instead of smooth L1, and the DIoU Soft-NMS mechanism was proposed to alleviate the misdetection in congested cabins. Further, comparison experiments and ablation experiments were performed on the auxiliary equipment in a marine engine room (AEMER) dataset to validate the efficacy of these strategies on the model performance boost. Specifically, our model can correctly detect 93.44% of coolers, 100.00% of diesel engines, 60.26% of meters, 95.30% of pumps, 55.01% of reservoirs, 97.68% of oil separators, and 74.37% of valves in a practical cabin. [ABSTRACT FROM AUTHOR]

It is important to accurately classify the defects in hot rolled steel strip since the detection of defects in hot rolled steel strip is closely related to the quality of the final product. The lack of actual hot-rolled strip defect data sets currently limits further research on the classification of hot-rolled strip defects to some extent. In real production, the convolutional neural network (CNN)-based algorithm has some difficulties, for example, the algorithm is not particularly accurate in classifying some uncommon defects. Therefore, further research is needed on how to apply deep learning to the actual detection of defects on the surface of hot rolled steel strip. In this paper, we proposed a hot rolled steel strip defect dataset called Xsteel surface defect dataset (X-SDD) which contains seven typical types of hot rolled strip defects with a total of 1360 defect images. Compared with the six defect types of the commonly used NEU surface defect database (NEU-CLS), our proposed X-SDD contains more types. Then, we adopt the newly proposed RepVGG algorithm and combine it with the spatial attention (SA) mechanism to verify the effect on the X-SDD. Finally, we apply multiple algorithms to test on our proposed X-SDD to provide the corresponding benchmarks. The test results show that our algorithm achieves an accuracy of 95.10% on the testset, which exceeds other comparable algorithms by a large margin. Meanwhile, our algorithm achieves the best results in Macro-Precision, Macro-Recall and Macro-F1-score metrics. [ABSTRACT FROM AUTHOR]