Search

Due to the limited availability of power defect data， most current defect detection methods are unable to accurately detect power system anomalies. To overcome this challenge， a few-shot image generation method is employed. Building upon the improved local-fusion generative adversarial network （LoFGAN）， a context-aware few-shot image generator is designed to enhance the defect detection network’s capability to extract detailed features. A regularization loss based on LC-divergence is introduced to optimize the training effectiveness of the image generation model on limited datasets. Experimental results reveal that the few-shot image generation method can generate effective and diverse defect data for power scenarios. The proposed model can address the issue of data unavailability in power defect scenarios.

The detection of traffic congestion is now realized mostly by human monitoring and sensor monitoring. However， such detection devices are deficient in road-occupied electric power construction. To meet the needs of low equipment dependency and high accuracy of congestion detection in the road-occupied electric power construction， a detection method based on video data is proposed， which uses neural networks to extract features from video data and determine whether there is traffic congestion. In response to data deficiency in the road-occupied electric power construction， the generalization of the network is improved by making full use of the generic traffic scene dataset， and the adaptive learning method based on domain adversarial neural networks （DANN） is used to reduce the differential performance of two data domains in the feature extraction network. Semi-supervised learning （SSL） is proposed to reduce the manual labeling workload. The experimental results show that the proposed method can achieve an accuracy of 93.2% in traffic congestion detection and recognition in road-occupied electric power construction and has high application value.

The existing classical object detection algorithm cannot satisfactorily detect abnormal behavior of operators at smart construction sites for its low accuracy. To this end， a two-stage detection algorithm based on the FCOS （full convolutional single-stage target detection） is proposed to detect abnormal behavior at smart construction sites. The algorithm， mainly consisting of two cascaded networks， first identifies and locates operators and abnormal behavior markers by the FCOS， and then uses the MLP （multilayer perceptron） to detect and classify abnormal behavior. Finally， the 12，977 sample images collected from the relevant project sites are used as the data set to experimentally validate the detection algorithm. The results show that the algorithm performs well in the detection of all kinds of abnormal behavior， and it has a clear advantage in the deployment and application of practical projects due to its excellent real-time detection， less complex computation， and fewer model parameters.

There are two problems in flame detection in electric power scenarios based on deep learning： 1）In contrast to the clear boundaries of general objects， the flame boundary is much hazy； 2）The deep learning-based methods require a large amount of labeled data for training； however， there is a severe shortage of large-scale flame detection data in electric power scenarios. In response to the first problem， this paper improves the traditional YOLOv5 detector to capture the uncertainty of the frame through Gaussian modeling. For the second question， this paper proposes a two-stage training method based on migration learning， which requires only a small number of flame data of the power scenarios for high-precision flame detector training. For verifying the effectiveness of the proposed method， accessible data on the internet and private data are used for contrast experiment. The results show that the proposed method is of great adaptability.

In view of electric power construction site with low level of informationization， lack of intelligent management， and small supervision areas， the paper proposes a smart monitoring method for electric power construction site areas based on target correlation method. Given the cumbersome positioning process of the targets to be measured （person， car） in the existing method and the inability to fully utilize the environment-related information， the correlation fusion module is introduced to improve the information utilization rate of the relative position of the targets to be tested. Multi-target positioning is carried out using the same network model to reduce the complexity of positioning and improve positioning efficiency. Experiments are carried out using real images of the construction site. The results show that the detection accuracy of the proposed method for the target to be measured in the construction site area is up to 99.2%， which is of high engineering application value.

In the existing electric construction scenario， the intelligent supervision scheme for wearing workwear focuses on safety helmets， and there are few schemes for the overall workwear compliance of operators. In view of the more detailed requirements of workers′ workwear compliance supervision， a two-stage compliance detection algorithm is proposed， which includes the personnel positioning stage and the human body area workwear compliance detection stage. Given the complex working posture of field personnel， a personnel positioning algorithm integrating FPN （feature pyramid network） and Guided Anchor based on Faster R-CNN is proposed. On the data set composed of more than 15,000 on-site collected samples， a personnel positioning accuracy of 91.11% is obtained， 6.0% higher than that of the ordinary Faster R-CNN scheme. The two-stage detection algorithm achieves an accuracy of 92.9% in the human body area workwear detection task， which is 11.4% higher than the single-stage Faster R-CNN scheme.

The discrepancy in data distribution between training and testing scenarios, as well as the inductive bias of convolutional neural networks towards image styles, reduces the model's generalization ability. Many unsupervised domain generalization methods based on feature decoupling suffer from an initial neglect of explicit decoupling of content and style features, resulting in content features that still contain considerable redundant information, thereby restricting improvements in generalization capability. To tackle this problem, this paper optimizes the learning process of domain‐invariant (content) features into an information compression issue, minimizing redundancy in content features. Furthermore, to enhance decoupled learning, this paper introduces innovative cross‐domain loss functions and image reconstruction modules that explicitly decouple and merge content and style across different domains. Extensive experiments demonstrate the method's significant enhancements over recent cutting‐edge approaches. [ABSTRACT FROM AUTHOR]

Whether percutaneous coronary intervention for chronic total occlusion (CTO-PCI) in diabetic patients offers more benefits compared with initial medical therapy (CTO-MT) is unclear. In this study, diabetic patients with one CTO (clinical manifestations: stable angina or silent ischemia) were enrolled. Consecutively, enrolled patients (n = 1605) were assigned to different groups: CTO-PCI (1044 [65.0%]) and initial CTO-MT (561 [35%]). After a median follow-up of 44 months, CTO-PCI tended to be superior to initial CTO-MT in major adverse cardiovascular events (adjusted hazard-ratio [aHR]:.81, 95% conference-interval:.65–1.02) and significantly superior in cardiac death (aHR:.58 [.39–.87]) and all-cause death (aHR:.678[.473–.970]). Such superiority mainly attributed to a successful CTO-PCI. CTO-PCI tended to be performed in patients with younger age, good collaterals, left anterior descending branch CTO, and right coronary artery CTO. While, those with left circumflex CTO and severe clinical/angiographic situations were more likely to be assigned to initial CTO-MT. However, none of these variables influenced the benefits of CTO-PCI. Thus, we concluded that for diabetic patients with stable CTO, CTO-PCI (mainly successful CTO-PCI) offered patients survival benefits over initial CTO-MT. These benefits were consistent regardless of clinical/angiographic characteristics. [ABSTRACT FROM AUTHOR]