Backbone Network Research Articles

A Fault Diagnosis Model for Insulator Instance Segmentation in Transmission Lines Based on ELD-YOLOv8

Abstract A novel ELD-YOLOv8n transmission line insulator instance segmentation fault diagnosis model is proposed to accurately and meticulously segment every fault of the transmission line insulator. Firstly, an innovative and efficient lightweight downsampling module (ELD) was proposed, Efficient-Lightweight downsampling, This module is used to replace the standard downsampling unit in the model, which not only reduces the number of model parameters, but also enhances the feature extraction ability of the model; Then, a lightweight CARAFE module was used to replace the upsampling of the model, optimizing the upsampling process and reducing the number of parameters; Finally, CGAFsion is used to fuse the features extracted from the backbone network with the head features, effectively compensating for the information loss caused by the convolution process. The experimental results show that the improved model proposed in this study mAP@50 The indicator reached 86.2%. The effectiveness of the im-provement and significant instance segmentation fault detection capability have been demonstrated through ablation experiments. This study provides a new technical path for fault diagnosis of insulator instance segmentation in transmission lines.

Adaptive spatial-channel feature fusion and self-calibrated convolution for early maize seedlings counting in UAV images

Accurate counting of crop plants is essential for agricultural science, particularly for yield forecasting, field management, and experimental studies. Traditional methods are labor-intensive and prone to errors. Unmanned Aerial Vehicle (UAV) technology offers a promising alternative; however, varying UAV altitudes can impact image quality, leading to blurred features and reduced accuracy in early maize seedling counts. To address these challenges, we developed RC-Dino, a deep learning methodology based on DINO, specifically designed to enhance the precision of seedling counts from UAV-acquired images. RC-Dino introduces two innovative components: a novel self-calibrating convolutional layer named RSCconv and an adaptive spatial feature fusion module called ASCFF. The RSCconv layer improves the representation of early maize seedlings compared to non-seedling elements within feature maps by calibrating spatial domain features. The ASCFF module enhances the discriminability of early maize seedlings by adaptively fusing feature maps extracted from different layers of the backbone network. Additionally, transfer learning was employed to integrate pre-trained weights with RSCconv, facilitating faster convergence and improved accuracy. The efficacy of our approach was validated using the Early Maize Seedlings Dataset (EMSD), comprising 1,233 annotated images of early maize seedlings, totaling 83,404 individual annotations. Testing on this dataset demonstrated that RC-Dino outperformed existing models, including DINO, Faster R-CNN, RetinaNet, YOLOX, and Deformable DETR. Specifically, RC-Dino achieved improvements of 16.29% in Average Precision (AP) and 8.19% in Recall compared to the DINO model. Our method also exhibited superior coefficient of determination (R²) values across different datasets for seedling counting. By integrating RSCconv and ASCFF into other detection frameworks such as Faster R-CNN, RetinaNet, and Deformable DETR, we observed enhanced detection and counting accuracy, further validating the effectiveness of our proposed method. These advancements make RC-Dino particularly suitable for accurate early maize seedling counting in the field. The source code for RSCconv and ASCFF is publicly available at https://github.com/collapser-AI/RC-Dino, promoting further research and practical applications.

Evaluating sowing uniformity in hybrid rice using image processing and the OEW-YOLOv8n network

Sowing uniformity is an important evaluation indicator of mechanical sowing quality. In order to achieve accurate evaluation of sowing uniformity in hybrid rice mechanical sowing, this study takes the seeds in a seedling tray of hybrid rice blanket-seedling nursing as the research object and proposes a method for evaluating sowing uniformity by combining image processing methods and the ODConv_C2f-ECA-WIoU-YOLOv8n (OEW-YOLOv8n) network. Firstly, image processing methods are used to segment seed image and obtain seed grids. Next, an improved model named OEW-YOLOv8n based on YOLOv8n is proposed to identify the number of seeds in a unit seed grid. The improved strategies include the following: (1) Replacing the Conv module in the Bottleneck of C2f modules with the Omni-Dimensional Dynamic Convolution (ODConv) module, where C2f modules are located at the connection between the Backbone and Neck. This improvement can enhance the feature extraction ability of the Backbone network, as the new modules can fully utilize the information of all dimensions of the convolutional kernel. (2) An Efficient Channel Attention (ECA) module is added to the Neck for improving the network’s capability to extract deep semantic feature information of the detection target. (3) In the Bbox module of the prediction head, the Complete Intersection over Union (CIoU) loss function is replaced by the Weighted Intersection over Union version 3 (WIoUv3) loss function to improve the convergence speed of the bounding box loss function and reduce the convergence value of the loss function. The results show that the mean average precision (mAP) of the OEW-YOLOv8n network reaches 98.6%. Compared to the original model, the mAP improved by 2.5%. Compared to the advanced object detection algorithms such as Faster-RCNN, SSD, YOLOv4, YOLOv5s YOLOv7-tiny, and YOLOv10s, the mAP of the new network increased by 5.2%, 7.8%, 4.9%, 2.8% 2.9%, and 3.3%, respectively. Finally, the actual evaluation experiment showed that the test error is from −2.43% to 2.92%, indicating that the improved network demonstrates excellent estimation accuracy. The research results can provide support for the mechanized sowing quality detection of hybrid rice and the intelligent research of rice seeder.

YOLO-fall: a YOLO-based fall detection model with high precision, shrunk size, and low latency

Abstract According to recent research statistics, falling has become an important factor affecting the health and safety of the elderly. To reduce the computational cost of hardware and meet the demand for real-time fall detection, we propose a lightweight fall detection network called YOLO-fall oriented for mobile and small edge computing devices. We have made the following improvements based on you only look once (YOLO). First, the backbone network is designed to be lightweight. Then, the convolution module is reparameterized and the C3 structure is improved to ensure the balance between speed and accuracy. Finally, a 5 × 5 depth convolution is added to the detection head to improve the detection ability of large targets. The proposed YOLO-fall is trained and validated on the E-FPDS public dataset and achieves a 78.4% mean average precision (mAP) with 2.45 M parameters and 12.2 GFLOPs. Compared with YOLOv5s, YOLO-fall has a 6.1% improvement in mAP and a 65.1% reduction in parameters. Although Yolov9s has a higher mAP of 82.9%, YOLO-fall reduces the parameters and calculation quantities by 74.8 and 69.2%, respectively. Therefore, the proposed YOLO-fall has the potential to accurately perform real-time fall detection on mobile and small edge computing devices.

Enhancing Food Image Recognition by Multi-Level Fusion and the Attention Mechanism

As a pivotal area of research in the field of computer vision, the technology for food identification has become indispensable across diverse domains including dietary nutrition monitoring, intelligent service provision in restaurants, and ensuring quality control within the food industry. However, recognizing food images falls within the domain of Fine-Grained Visual Classification (FGVC), which presents challenges such as inter-class similarity, intra-class variability, and the complexity of capturing intricate local features. Researchers have primarily focused on deep information in deep convolutional neural networks for fine-grained visual classification, often neglecting shallow and detailed information. Taking these factors into account, we propose a Multi-level Attention Feature Fusion Network (MAF-Net). Specifically, we use feature maps generated by the Convolutional Neural Networks (CNNs) backbone network at different stages as inputs. We apply a self-attention mechanism to identify local features on these feature maps and then stack them together. The feature vectors obtained through the attention mechanism are then integrated with the original input to enhance data augmentation. Simultaneously, to capture as many local features as possible, we encourage multi-scale features to concentrate on distinct local regions at each stage by maximizing the Kullback-Leibler Divergence (KL-divergence) between the different stages. Additionally, we present a novel approach called subclass center loss (SCloss) to implement label smoothing, minimize intra-class feature distribution differences, and enhance the model’s generalization capability. Experiments conducted on three food image datasets—CETH Food-101, Vireo Food-172, and UEC Food-100—demonstrated the superiority of the proposed model. The model achieved Top-1 accuracies of 90.22%, 89.86%, and 90.61% on CETH Food-101, Vireo Food-172, and UEC Food-100, respectively. Notably, our method not only outperformed other methods in terms of the Top-5 accuracy of Vireo Food-172 but also achieved the highest performance in the Top-1 accuracies of UEC Food-100.

Real-Time High Dynamic Equalization Industrial Imaging Enhancement Based on Fully Convolutional Network

Severe reflections on the surfaces of smooth objects can result in low dynamic range and uneven illumination in images, which negatively impacts downstream tasks such as defect detection and QR code recognition on images of smooth workpieces. Consequently, this paper proposes a novel approach to real-time high dynamic equalization imaging based on a fully convolutional network, termed Multi-exposure Image Fusion with Multi-dimensional Attention Mechanism and Training Storage Units (MEF-AT). Specifically, this paper innovatively proposes using training storage units, which utilize intermediate results during network training as auxiliary images, to remove uneven illumination and enhance image dynamic range effectively. Furthermore, by integrating a multi-dimensional attention mechanism into the backbone network, the model can more efficiently extract and utilize critical image information. Additionally, this paper introduces a Deep Guided Filter (DGF) with learnable parameters, which upsample the weight maps generated by the network, thus better adapting to complex industrial scenarios and producing higher quality fused images. An image evaluation metric assessing the lighting uniformity is introduced to thoroughly evaluate the proposed method’s performance. Given the lack of an MEF dataset for smooth workpieces, this paper collects a new dataset for multi-exposure fusion tasks on metallic workpieces. Our method takes less than 4 ms to run four 2K images on a GPU 3090. Both qualitative and quantitative experimental results demonstrate our method’s superior comprehensive performance in proprietary industrial and public datasets.

Research on power plant security issues monitoring and fault detection using attention based LSTM model

OverviewFor Photo Voltaic (PV) arrays and Wind systems to operate as efficiently and effectively as possible, fault detection is essential. It is possible to improve the safety of renewable energy systems and guarantee that the service will continue uninterrupted if problem detection and diagnostics are performed in a timely and accurate manner. In general, wind power is one of the three major renewable energy sources, along with solar power and hydropower. Wind power is well distributed around the world, making it suitable to be exploited in human activities for the general welfare of society.ObjectivesA prototype security situational awareness system applicable to the power data communication network service and traffic model should be developed. This will help to successfully enhance the security and service quality of the power data communication network, effectively cope with network security threats in the new environment, and ensure the security of the power plant network access. The traffic of the main network of the existing data communication network will be combined and analyzed, and NQA traffic management algorithms will be studied and proposed. These actions will improve the SLA hierarchical service capability, the service quality of the core services carried by the backbone network, and strengthen the security capability of the new energy power plant communication network access system.MethodologyFor the purpose of this investigation, an attention-based long short-term memory (Att-LSTM) model was used for the categorization of time series actual data. The approach that has been developed is able to identify defects in photovoltaic arrays and inverters, which offers a dependable option for improving the efficiency and dependability of solar energy systems. For the purpose of evaluating the proposed method, a real-world solar energy dataset is used.ResultsThe findings acquired from this evaluation are compared to the results received from existing detection approaches such as Cryptography, Intrusion Detection System (IDS) methods, and Network Defense Schemes. The results obtained demonstrate that the suggested method surpasses current fault detection techniques, providing greater accuracy and better performance.

A Lightweight Dynamically Enhanced Network for Wildfire Smoke Detection in Transmission Line Channels

In view of the problems that mean that existing detection networks are not effective in detecting dynamic targets such as wildfire smoke, a lightweight dynamically enhanced transmission line channel wildfire smoke detection network LDENet is proposed. Firstly, a Dynamic Lightweight Conv Module (DLCM) is devised within the backbone network of YOLOv8 to enhance the perception of flames and smoke through dynamic convolution. Then, the Ghost Module is used to lightweight the model. DLCM reduces the number of model parameters and improves the accuracy of wildfire smoke detection. Then, the DySample upsampling operator is used in the upsampling part to make the image generation more accurate with very few parameters. Finally, in the course of the training process, the loss function is improved. EMASlideLoss is used to improve detection ability for small targets, and the Shape-IoU loss function is used to optimize the shape of wildfires and smoke. Experiments are conducted on wildfire and smoke datasets, and the final mAP50 is 86.6%, which is 1.5% higher than YOLOv8, and the number of parameters is decreased by 29.7%. The experimental findings demonstrate that LDENet is capable of effectively detecting wildfire smoke and ensuring the safety of transmission line corridors.

Interactive segmentation for accurately isolating metastatic lesions from low-resolution, large-size bone scintigrams.

Bone is a common site for the metastasis of malignant tumors, and Single Photon Emission Computed Tomography (SPECT) is widely used to detect these metastases. Accurate delineation of metastatic bone lesions in SPECT images is essential for developing treatment plans. However, current clinical practices rely on manual delineation by physicians, which is prone to variability and subjective interpretation. While computer-aided diagnosis (CAD) systems have the potential to improve diagnostic efficiency, fully automated segmentation approaches frequently suffer from high false positive rates, limiting their clinical utility.
Approach: This study proposes an interactive segmentation framework for SPECT images, leveraging the deep convolutional neural networks (DCNN) to enhance segmentation accuracy. The proposed framework incorporates a U-shaped backbone network that effectively addresses inter-patient variability, along with an interactive attention module that enhances feature extraction in densely packed bone regions.
Main results: Extensive experiments using clinical data validate the effectiveness of the proposed framework. Furthermore, a prototype tool was developed based on this framework to assist in the clinical segmentation of metastatic bone lesions and to support the creation of a large-scale dataset for bone metastasis segmentation.
Significance: In this study, we proposed an interactive segmentation framework for metastatic lesions in bone scintigraphy to address the challenging task of labeling low-resolution, large-size SPECT bone scans. The experimental results show that the model can effectively segment the bone metastases of lung cancer interactively. In addition, the prototype tool developed based on the model has certain clinical application value.

Dual-prompt complementary fusion network for RGBT tracking

RGBT target tracking is a significant downstream task in the field of object tracking. However, compared to visible light target tracking, RGBT target tracking faces the challenge of smaller datasets, making it difficult to achieve performance levels comparable to those achieved in visible light target tracking. To address how to effectively combine the complementary characteristics of visible and thermal modalities, as well as how to fully leverage the superior performance of models trained on visible light target tracking tasks, while also aiming for lower computational costs and higher tracking effectiveness, a dual-prompt complementary fusion strategy for an RGBT tracking network is proposed. Drawing on the concept of prompt learning, this network aims to extend the efficient performance of visible light target tracking to the RGBT target tracking domain. In its implementation, the prompt module inputs both visible and thermal modality information as dual prompts into the backbone network, where the network utilizes these prompts to generate new, enriched prompt information at each layer. Subsequently, an information enhancement fusion module enhances the acquired prompt information and refeeds it into the backbone network, aiming to improve the tracking accuracy and robustness. Experimental results on GTOT, RGBT234 and LasHeR datasets show that the tracking accuracy (PR) and success rate (SR) of the network reach 93.1%/76.8%, 84.4%/62.4% and 66.8%/53.8%, respectively, which is improved compared with the current mainstream RGBT target tracking network, which verifies the effectiveness of the network.

DCFA-YOLO: A Dual-Channel Cross-Feature-Fusion Attention YOLO Network for Cherry Tomato Bunch Detection

To better utilize multimodal information for agriculture applications, this paper proposes a cherry tomato bunch detection network using dual-channel cross-feature fusion. It aims to improve detection performance by employing the complementary information of color and depth images. Using the existing YOLOv8_n as the baseline framework, it incorporates a dual-channel cross-fusion attention mechanism for multimodal feature extraction and fusion. In the backbone network, a ShuffleNetV2 unit is adopted to optimize the efficiency of initial feature extraction. During the feature fusion stage, two modules are introduced by using re-parameterization, dynamic weighting, and efficient concatenation to strengthen the representation of multimodal information. Meanwhile, the CBAM mechanism is integrated at different feature extraction stages, combined with the improved SPPF_CBAM module, to effectively enhance the focus and representation of critical features. Experimental results using a dataset obtained from a commercial greenhouse demonstrate that DCFA-YOLO excels in cherry tomato bunch detection, achieving an mAP50 of 96.5%, a significant improvement over the baseline model, while drastically reducing computational complexity. Furthermore, comparisons with other state-of-the-art YOLO and other object detection models validate its detection performance. This provides an efficient solution for multimodal fusion for real-time fruit detection in the context of robotic harvesting, running at 52fps on a regular computer.

Neural Network for Underwater Fish Image Segmentation Using an Enhanced Feature Pyramid Convolutional Architecture

Underwater fish image segmentation is a crucial technique in marine fish monitoring. However, typical underwater fish images often suffer from issues such as color distortion, low contrast, and blurriness, primarily due to the complex and dynamic nature of the marine environment. To enhance the accuracy of underwater fish image segmentation, this paper introduces an innovative neural network model that combines the attention mechanism with a feature pyramid module. After the backbone network processes the input image through convolution, the data pass through the enhanced feature pyramid module, where it is iteratively processed by multiple weighted branches. Unlike conventional methods, the multi-scale feature extraction module that we designed not only improves the extraction of high-level semantic features but also optimizes the distribution of low-level shape feature weights through the synergistic interactions of the branches, all while preserving the inherent properties of the image. This novel architecture significantly boosts segmentation accuracy, offering a new solution for fish image segmentation tasks. To further enhance the model’s robustness, the Mix-up and CutMix data augmentation techniques were employed. The model was validated using the Fish4Knowledge dataset, and the experimental results demonstrate that the model achieves a Mean Intersection over Union (MIoU) of 95.1%, with improvements of 1.3%, 1.5%, and 1.7% in the MIoU, Mean Pixel Accuracy (PA), and F1 score, respectively, compared to traditional segmentation methods. Additionally, a real fish image dataset captured in deep-sea environments was constructed to verify the practical applicability of the proposed algorithm.

An automatic cervical cell classification model based on improved DenseNet121

The cervical cell classification technique can determine the degree of cellular abnormality and pathological condition, which can help doctors to detect the risk of cervical cancer at an early stage and improve the cure and survival rates of cervical cancer patients. Addressing the issue of low accuracy in cervical cell classification, a deep convolutional neural network A2SDNet121 is proposed. A2SDNet121 takes DenseNet121 as the backbone network. Firstly, the SE module is embedded in DenseNet121 to increase the model’s focus on the nucleus region, which contains important diagnostic information, and reduce the focus on redundant information. Secondly, the sizes of the convolutional kernel and pooling window of the Stem layer are adjusted to adapt to the characteristics of the cervical cell images, so that the model can extract the local detailed information more effectively. Finally, the Atrous Dense Block (ADB) is constructed, and four ADB modules are integrated into DenseNet121 to enable the model to acquire global and local salient feature information. The accuracy of A2SDNet121 for two and seven-classification tasks on the Herlev dataset is 99.75% and 99.14%, respectively. The accuracy for two, three, and five-classification tasks on the SIPaKMeD dataset reaches 99.55%, 99.75% and 99.22%, respectively. Compared with other state-of-the-art algorithms, the A2SDNet121 model performs better in the multi-classification task of cervical cells, which can significantly improve the accuracy and efficiency of cervical cancer screening.

An Asymmetric Selective Kernel Network for Drone-Based Vehicle Detection to Build a High-Accuracy Vehicle Trajectory Dataset

To improve the detection accuracy of the drone-based oriented vehicle object detection network and establish high-accuracy vehicle trajectory datasets, we present a freeway on-ramp vehicle (FRVehicle) detection dataset with oriented bounding box annotations for vehicles in freeway on-ramp scenes from drone videos. Based on this dataset, we analyzed the dimension and angle distribution patterns of road vehicle object oriented bounding boxes and designed an Asymmetric Selective Kernel Network. This algorithm dynamically adjusts the receptive field of the backbone network’s feature extraction to accommodate the detection requirements for vehicles of different sizes. Additionally, we estimate vehicle heights with high-precision object detection results, further enhancing the accuracy of the vehicle trajectory. Comparative experimental results demonstrate that the proposed Asymmetric Selective Kernel Network achieved varying degrees of improvement in detection accuracy on both the FRVehicle dataset and DroneVehicle dataset compared to the symmetric selective kernel network in most scenarios, validating the effectiveness of the method.

FldtMatch: Improving Unbalanced Data Classification via Deep Semi-Supervised Learning with Self-Adaptive Dynamic Threshold

Among the many methods of deep semi-supervised learning (DSSL), the holistic method combines ideas from other methods, such as consistency regularization and pseudo-labeling, with great success. This method typically introduces a threshold to utilize unlabeled data. If the highest predictive value from unlabeled data exceeds the threshold, the associated class is designated as the data’s pseudo-label. However, current methods utilize fixed or dynamic thresholds, disregarding the varying learning difficulties across categories in unbalanced datasets. To overcome these issues, in this paper, we first designed Cumulative Effective Labeling (CEL) to reflect a particular class’s learning difficulty. This approach differs from previous methods because it uses effective pseudo-labels and ground truth, collectively influencing the model’s capacity to acquire category knowledge. In addition, based on CEL, we propose a simple but effective way to compute the threshold, Self-adaptive Dynamic Threshold (SDT). It requires a single hyperparameter to adjust to various scenarios, eliminating the necessity for a unique threshold modification approach for each case. SDT utilizes a clever mapping function that can solve the problem of differential learning difficulty of various categories in an unbalanced image dataset that adversely affects dynamic thresholding. Finally, we propose a deep semi-supervised method with SDT called FldtMatch. Through theoretical analysis and extensive experiments, we have fully proven that FldtMatch can overcome the negative impact of unbalanced data. Regardless of the choice of the backbone network, our method achieves the best results on multiple datasets. The maximum improvement of the macro F1-Score metric is about 5.6% in DFUC2021 and 2.2% in ISIC2018.

SED-YOLO based multi-scale attention for small object detection in remote sensing

Object detection is crucial for remote sensing image processing, yet the detection of small objects remains highly challenging due to factors such as image noise and cluttered backgrounds. In response to this challenge, this paper proposes an improved network, named SED-YOLO, based on YOLOv5s. Firstly, we leverage Switchable Atrous Convolution (SAC) to replace the standard convolutions in the original C3 modules of the backbone network, thereby enhancing feature extraction capabilities and adaptability. Additionally, we introduce the Efficient Multi-Scale Attention(EMA) mechanism at the end of the backbone network to enable efficient multi-scale feature learning, which reduces computational costs while preserving crucial information. In the Neck section, an adaptive Concat method is designed to dynamically adjust the feature fusion strategy according to image content and object characteristics, strengthening the model’s ability to handle diverse objects. Lastly, the three-scale feature detection head is expanded to four by adding a small object detection layer, and incorporating the Dynamic Head(DyHead) module. This enhances the detection head’s expressive power by dynamically adjusting attention weights in feature maps. Experimental results demonstrate that this improved network achieves an mean Average Precision (mAP) of 71.6% on the DOTA dataset, surpassing the original YOLOv5s by 2.4%, effectively improving the accuracy of small object detection.

Research on Vehicle Detection Based on Improved YOLOv8 Network

The key to ensuring the safe obstacle avoidance function of autonomous driving systems lies in the use of extremely accurate vehicle recognition techniques. However, the variability of the actual road environment and the diverse characteristics of vehicles and pedestrians together constitute a huge obstacle to improving detection accuracy, posing a serious challenge to the realization of this goal. To address the above issues, this paper proposes an improved YOLOv8 vehicle detection method. Specifically, taking the YOLOv8n-seg model as the base model, firstly, the FasterNet network is used to replace the backbone network to achieve the purpose of reducing the computational complexity and memory while improving the detection accuracy and speed; secondly, the feature enhancement is achieved by adding the attention mechanism CBAM to the Neck; and lastly, the loss function CIoU is modified to WIoU, which optimizes the detection box localization while improving the segmentation accuracy. The results show that the improved model achieves 98.3%, 89.1% and 88.4% detection accuracy for car, Person and Motorcycle. Compared with the pre-improvement and YOLOv9 models in six metrics such as Precision.

End-to-end Identification of Autoregressive with Exogenous Input (ARX) Models Using Neural Networks

Traditional parametric system identification methods usually rely on apriori knowledge of the targeted system, which may not always be available, especially for complex systems. Although neural networks (NNs) have been increasingly adopted in system identification, most studies have failed to derive interpretable parametric models for further analysis. In this paper, we propose a novel end-to-end autoregressive with exogenous input (ARX) model identification framework using NNs. An order-wise neural network structure is introduced and trained using a multitask learning approach to simultaneously identify both the model terms and coefficients of the ARX model. Through testing with various neural network backbones and training data sizes in different scenarios, we empirically demonstrate that the proposed framework can effectively identify an arbitrary stable ARX model with finite simulation training data. This study opens up a new research opportunity for parametric system identification by harnessing the power of deep learning.

An Improved Mask2Former-HRNet Method for Insulator Defect Detection

To solve the problem of scale variation in insulator images captured by drones, caused by the lack of control over angle and distance, which makes it hard to detect subtle defects, this paper proposes an instance segmentation method based on an improved Mask2Former-HRNet model for precise localization and defect detection of transmission line insulators. First, a mask-guided and matching component is added to Mask2Former to reduce the misjudgment rate of insulator defects by including noisy label masks. Second, the HRNet backbone network is used to better capture the spatial and shape information of insulators, as it has a stronger feature transfer ability. Deformable convolutions are introduced to handle deformation issues caused by varying angles in insulator images. Then, an attention mechanism is added to focus on key content, improving the network’s attention to crucial information. Finally, experimental results on defect detection of transmission line insulator images captured by drones show that the proposed method increases the detection accuracy by 8.41% and reduces the misjudgment rate by 4.11%. Comparative experiments indicate that the proposed method outperforms existing methods in several evaluation metrics.

Research on mechanical part recognition method based on improved mask R-CNN instance segmentation

Aiming at the problem of poor part recognition due to mutual occlusion between parts and the influence of different postures in the assembly scene, we propose an improved Mask R-CNN-based part recognition method for complex scenes. Firstly, the ResNet101 network is used to enhance the feature extraction capability of the network and improve the part recognition effect; secondly, the normalization layer of the backbone network is replaced to reduce the effect of batch size on the feature extraction of the model; lastly, the feature pyramid network structure is improved to enhance the transfer efficiency between the high and low layers of the network, and to enhance the capability of the feature capture; through the experiments on the homemade dataset, the average detection accuracy of this method is 4.7 % higher than that of the original Mask R-CNN. Through the experiments on the homemade dataset, it is found that compared with the original Mask R-CNN, the average detection accuracy of the method is improved by 4.7 %.The optimized network model proposed in this paper can improve the accuracy of part recognition, realize the accurate detection of parts in the complex environment such as stacking, occlusion and so on, and provide a solution for the recognition of parts in the complex environment.