The high‐quality 3D model is the key to digital twin and visualization. Here, the neural radiation field (NeRF) is applied to greatly improve the speed and accuracy of 3D reconstruction of large‐scale substation equipment. The incremental SFM algorithm is used to accurately position the camera in open scenes, and the frame extraction experiment indicates that the single angle increment should not be greater than 8°. The reconstruction test is carried out on the power transformer and high‐voltage experimental equipment following the circular shooting track for instant‐NGP (a variant of NeRF). Compared with the traditional 3D reconstruction method, the speed of reconstruction is accelerated by 23–55 times, and the fidelity is improved by 2–3 times. Also, the deployment scheme of NeRF in the digital twin technology of large‐scale substation equipment is proposed. The experimental results verify the effectiveness and feasibility of NeRF in practical application.

In order to facilitate the digital transformation of conventional power grid, constructing the digital twin of power equipment is crucial. The primary task of the digital transformation is to build the 3D model of power equipment. However, existing methods cannot achieve digital twin modeling with high fidelity and efficiency. Therefore, a 3D reconstruction method for heterogenous data based on improved Neural Radiation Fields (NeRF) is proposed. By utilizing the infrared-visible image sequence, the spatial domain consistency of the heterogenous data is preserved by the image registration algorithm improved based on restricted matching region strategy (RMRS). Then, in order to focus on the characteristics of the equipment, a deep image matting network (DIM) is designed to eliminate the interference of complex background in the image. The pose of the camera is recovered from the visible images of the power equipment based on Structure From Motion (SFM). Finally, the NeRF is employed to complete the 3D reconstruction of the visible model and temperature field model of power equipment, enabling 3D fusion perception of the 2D heterogenous monitoring data. Experimental results show that this method can realize high-fidelity 3D reconstruction of power equipment within 30 s, which is superior to existing methods in terms of accuracy, modeling speed, automation degree, robustness and lightweight.

Insulators, as a key component of the power system, their low‐value defect detection is of great significance to ensure the safe and stable operation of the power system. However, traditional detection methods have many shortcomings in the face of a complex environment and small target recognition. To solve the above problems, this paper optimizes the small target and complex environment problems in the low‐value defect recognition of insulator infrared images, and proposes the STCE‐YOLO algorithm: based on YOLOv8, the deformable large kernel attention is used to improve the detection ability of small targets; then the cross‐modal contextual feature module is applied to Integrate the features of different scales to reduce the computation of the model. And the multiple attention mechanism improved to the third generation of variability convolution is used to detect the head to improve the accuracy of the algorithm's target localization. Finally, the SIoU loss function is employed to further enhance performance in complex scenes containing small targets. Experimental validation has shown that the STCE‐YOLO algorithm proposed in this paper achieves an average improvement of 7.64% in mAP compared to the original YOLOv8, with GFLOPs reduced from 8.1 to 7.7. This meets the requirements for identifying low‐value defects in small target insulators. Furthermore, ablation and comparative experiments have demonstrated the effectiveness and superiority of the proposed algorithm.

Edge devices are increasingly utilized for defect detection in power line inspection, necessitating algorithms that optimize model size and accuracy. This study introduces a lightweight detection method using the Optimized Feature Network with MobileOne‐FPN‐NASHead (OFN network) OFN network and distillation technique. The OFN incorporates a lightweight backbone, neural network search, re‐parametrization, and a feature pyramid to create a compact yet effective detection network. To enhance feature learning, a heterogeneous distillation approach is applied, leveraging a modified YOLOv8 as a teacher network. This modification includes an explicit visual centre for improved global and local information extraction, crucial for dense target detection in power line inspections. Additionally, the Minimize the points distance IoU (MPDloU) loss function is used to improve localization accuracy over the the Complete‐Intersection Over Union (CIoU) loss. Experimental results show a 1.1% mean Average Precision (mAP) increase for the enhanced YOLOv8 and a 70.2% mAP for the OFN network with 18.95 GFLOPs and 343 FPS, achieving a commendable balance between model efficiency and detection performance. The research underscores the viability of the OFN for edge computing in power line defect detection, highlighting the potential of innovative algorithmic structures in this application.

In the field of power equipment inspection, the aging condition of composite insulators is often determined by the detection of water repellency. However, the existing detection methods are difficult to effectively extract the water repellency level features in the complex background, and it is difficult to meet the real‐time requirements. Therefore, this paper proposes a MDC‐YOLO algorithm for water repellency detection and classification of composite insulators. The algorithm effectively combines the efficient convolutional network (EfficientNetV2), the deformable attention mechanism (DAttention), and the lightweight convolution (CSPPC), which significantly realizes the lightweight of the network, and at the same time improves the accuracy of the algorithm for the composite insulator umbrella skirt identification, water repellency class classification. Experimentally verified, the multilayer detection and classification YOLO algorithm of water repellency of composite insulators proposed in this paper, i.e., MDC‐YOLO algorithm, improves the detection accuracy of umbrella skirt by 5.19% and reduces the GFLOPs to 3.5, and improves the Top‐1 accuracy of water repellency grade classification by 4.654% and reduces the GFLOPs by 1.1. The results of the present research can be widely applied to The results of this study can be widely used in the fields of power equipment inspection, smart grid construction and insulator aging state assessment, which provides strong support for the technical development of related fields. It meets the requirements of hydrophobicity detection and classification of composite insulators, and proves the effectiveness and superiority of the algorithm proposed in this paper through ablation and comparison tests.

The axial temperature is often used to reflect the condition of insulators in research on the insulator state. Traditional manual extraction methods are cumbersome. To improve research efficiency, this paper proposes a method for extracting insulator axis problems in order to address the challenge of obtaining a large number of insulator axis temperatures quickly in commercial software. The Retinaface network is simplified to develope a streamlined version called S-Retinaface, which facilitates training with infrared images and enables seamless industrial deployment. This method harnesses the S-Retinaface network, tailored for infrared data, ensuring precise detection of insulator keypoints. From these, the insulator's axis line equation can be computed, leading to the precise mapping of axial temperature profiles. Experimental validation on a test dataset demonstrates the accuracy of S-Retinaface, reflected by a high Percentage of Correct Keypoints Area Under Curve of 0.864, and showcases a commendable similarity with manual extraction methods, as evidenced by an average Dynamic Time Warping distance of 2.1008. This technique not only elevates the efficiency of temperature data collection but also promises an impact on the field's operational dynamics, offering a rapid, dependable means to assess insulator integrity.