Bolt defect inspection is an important work in transmission line inspection. Due to the small size of bolts in the transmission line inspection images, existing algorithms are difficult to extract valuable features and achieve poor performance on bolt defect detection. This paper proposed an ultra-small bolt defect detection model(UBDDM) based on a deep convolutional neural network(DCNN), including an ultra-small object perception module(UOPM) and a local bolt detection module(LBDM). In this paper, UOPM is first constructed to realize coarse region recognition for the salient region of bolts in the inspection images, and the high-resolution image blocks are obtained from the original image according to the recognition results. Then, LBDM is constructed to intelligently identify the bolt defects from the high-resolution image blocks. Considering that the features of ultra-small targets are difficult to extract, feature extraction networks are constructed based on ResNet-50, and the hybrid attention mechanism and multi-scale feature fusion are introduced to further improve the network’s ability to extract shallow features. This method uses two-stage detection to realize end-to-end bolt defect detection but only needs to provide a single-stage target detection label, which greatly reduces the workload of data labeling. Experimental results show that the proposed method achieves excellent performance on bolt defect detection in inspection images.

Icing monitoring system within internet of things is developed to provide multimodal data for assessing the degree of icing on transmission lines. Nevertheless, existing methods relying solely on either sensor data or images demonstrate limited precision and inferior fault tolerance. This paper proposes a correlation-based multimodal feature fusion approach to integrate both sensor data and imaging data, thereby enhancing the monitoring of icing severity on transmission lines and addressing the aforementioned problems. The inherent correlation characteristics present in both sensor data and images, as well as the correlation between these two modalities are thoroughly analyzed, to gain a comprehensive understanding of icing characteristics in multimodal data. Specifically, the squeeze-excitation module along with convolutional neural networks are combined to extract the temporal and spatial correlation inherent in sensor data, as well as the spatial and channel correlation inherent in images. Subsequently, the covariance matrix is used to capture the correlation between these two modalities. Moreover, with the assistance of the weight determination structure, this correlation is mapped to the fusion weights. The four-stage training strategy is introduced to guide the network training. The essentiality of the staged training approach and the necessity of correlation extraction in perceiving icing severity are validated in experiments.

The safe approach distance detection of large construction equipment in substations is important to ensure the safety and stability of the power system, as well as to prevent equipment damage, power outages and other accidents. The current method is unable to intelligently distinguish construction equipment from power equipment and realize real-time safety approach distance detection. Therefore, this paper constructs a safety approach distance detection system for large-scale construction equipment in substations based on stereo vision and target detection, and realizes real-time high-precision safe approach distance detection between large-scale construction equipment and electric power equipment. Firstly, the system distinguishes construction equipment from power equipment using a GhostNet-based substation construction target detection model. Secondly, the system obtains spatial information regarding the operation scene using a lightweight stereo matching model based on channel attention, then calculates the spatial surface center of the target based on the spatial information and detection results, and finally calculates the safety approach distance between construction equipment and power equipment. Compared with MobileNetv3-YOLOv4, the map and the recall rate of the proposed method are improved by 13.1% and 29.0%; compared with the AnyNet stereo matching method, the proposed method decreases the end point error and 3 pixels error by 34.2% and 25.8%. The actual data show that the detection speed of the proposed method is 19.35 frames per second, and the mean absolute error is 0.942 m and the mean relative error is 3.802%. This method can accurately measure the safe approach distance in real time in real scenarios to guarantee the safety of personnel and equipment.

Wildfire detection in power transmission corridors is essential for providing timely warnings and ensuring the safe and stable operation of power lines. However, this task faces significant challenges due to the large number of smoke-like samples in the background, the complex and diverse target morphologies, and the difficulty of detecting small-scale smoke and flame objects. To address these issues, this paper proposed an improved Oriented R-CNN model enhanced with metric learning for wildfire detection in power transmission corridors. Specifically, a multi-center metric loss (MCM-Loss) module based on metric learning was introduced to enhance the model’s ability to differentiate features of similar targets, thereby improving the recognition accuracy in the presence of interference. Experimental results showed that the introduction of the MCM-Loss module increased the average precision (AP) for smoke targets by 2.7%. In addition, the group convolution-based network ResNeXt was adopted to replace the original backbone network ResNet, broadening the channel dimensions of the feature extraction network and enhancing the model’s capability to detect flame and smoke targets with diverse morphologies. This substitution led to a 0.6% improvement in mean average precision (mAP). Furthermore, an FPN-CARAFE module was designed by incorporating the content-aware up-sampling operator CARAFE, which improved multi-scale feature representation and significantly boosted performance in detecting small targets. In particular, the proposed FPN-CARAFE module improved the AP for fire targets by 8.1%. Experimental results demonstrated that the proposed model achieved superior performance in wildfire detection within power transmission corridors, achieving a mAP of 90.4% on the test dataset—an improvement of 6.4% over the baseline model. Compared with other commonly used object detection algorithms, the model developed in this study exhibited improved detection performance on the test dataset, offering research support for wildfire monitoring in power transmission corridors.

The inspection of power transmission lines using unmanned aerial vehicles primarily relies on object detection. However, the continuous emergence of new obstacle types necessitates frequent updates to detection models, leading to substantial retraining costs. To address this challenge, we propose a novel framework named IViT, which integrates incremental learning with a hybrid CNN-Transformer architecture for improved identification. We combined knowledge distillation with the elastic response selection distillation strategy to enhance detection performance for old classes and strengthen knowledge retention through star convolutional residual blocks constructed via element-wise multiplication. We designed a separable convolution aggregation block that integrates PConv with an attention mechanism, effectively merging global and local information to improve detection accuracy. Finally, we unified the two modules into a hybrid block. In the static detection task, IViT achieves a mAP of 55.3%, a mAP₅₀ of 83.6%, and a mAP₇₅ of 61.0%. For the incremental detection task, it attains a mAP of 57.8%, a mAP₅₀ of 79.7%, and a mAP₇₅ of 62.3%. Extensive experiments on the transmission line corridor external damage dataset and the INSPLAD dataset demonstrate that IViT exhibits outstanding detection performance compared to mainstream static object detection models and incremental object detection models.

Continuous monitoring of the minimum safety distance between construction machinery and energized bodies is essential during operations near energized equipment in substations. Existing methods mostly rely on fixed-view observation, online LiDAR, or rigid camera–LiDAR installation, leading to inflexible deployment, high extrinsic-maintenance cost, and insufficient metric consistency across viewpoints. To address these limitations, this paper proposes a safety-distance monitoring method based on cross-modal self-registration between monocular images and a pre-built LiDAR map. During online operation, only the current monocular image is used. Monocular depth estimation first generates a pseudo-point cloud, which is then registered with historical LiDAR point clouds to solve the camera pose and align the current observation with the map. Combined with target-boundary segmentation and prior energized-hazardous-region information, the method localizes key parts of construction machinery in 3D and computes the minimum safety distance to energized regions. Experiments show that the proposed method achieves a registration recall of 92.2%, a mean absolute error of 0.748 m, a maximum error of 0.822 m, and a single-frame latency of 180 ms. Notably, these results are achieved without real-time LiDAR input or on-site extrinsic recalibration. These results demonstrate the feasibility of the proposed framework in a representative substation scenario and indicate its potential for auxiliary online safety-distance monitoring.