Abstract:Power grids serve as a vital component in numerous industries, seamlessly delivering electrical energy to industrial processes and technologies, making their safe and reliable operation indispensable. However, powerlines can be hard to inspect due to difficult terrain or harsh climatic conditions. Therefore, unmanned aerial vehicles are increasingly deployed to inspect powerlines, resulting in a substantial stream of visual data which requires swift and accurate processing. Deep learning methods have become widely popular for this task, proving to be a valuable asset in fault detection. In particular, the detection of insulator defects is crucial for predicting powerline failures, since their malfunction can lead to transmission disruptions. It is therefore of great interest to continuously maintain and rigorously inspect insulator components. In this work we propose a novel pipeline to tackle this task. We utilize state-of-the-art object detection to detect and subsequently classify individual insulator anomalies. Our approach addresses dataset challenges such as imbalance and motion-blurred images through a fine-tuning methodology which allows us to alter the classification focus of the model by increasing the classification accuracy of anomalous insulators. In addition, we employ explainable-AI tools for precise localization and explanation of anomalies. This proposed method contributes to the field of anomaly detection, particularly vision-based industrial inspection and predictive maintenance. We significantly improve defect detection accuracy by up to 13%, while also offering a detailed analysis of model mis-classifications and localization quality, showcasing the potential of our method on real-world data.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the dataset imbalance issue in power - line insulator defect detection. Specifically, the paper focuses on how to effectively detect and classify insulator anomalies in highly imbalanced datasets (for example, the number of healthy insulator samples is far greater than that of damaged samples), especially those critical defects that may lead to power transmission interruptions. In addition, the paper also explores how to use Explainable Artificial Intelligence (XAI) tools to accurately locate and explain these anomalies. ### Main contributions of the paper 1. **Proposed a new insulator defect detection pipeline**: This pipeline can not only detect the entire insulator but also detect each insulator shell separately to reduce the interference of background features. 2. **Analyze the performance of each class through a re - weighting step and improve the performance of the worst class**: In response to the dataset imbalance problem, the method of retraining the classification head is adopted to improve the detection accuracy of defective classes. 3. **Proposed a new method to visualize and locate defects on insulator shells**: Using the Layer - wise Relevance Propagation (LRP) technique to provide pixel - level anomaly localization. 4. **Analyze model localization and study the problem of image misclassification**: Focus on examining the influence of image clarity and motion blur on fault detection. ### Method overview - **Two - stage detection strategy**: - First stage: Use the YOLOv8 network to detect insulators and generate bounding boxes. - Second stage: Crop the detected insulators from the original image, and then use another YOLOv8 model to detect individual insulator shells and perform classification. - **Fine - tuning method against dataset imbalance**: - Since the number of healthy shells far exceeds that of damaged shells, the model is biased towards the healthy category. To solve this problem, the author uses logistic regression to retrain the last layer and introduces class weights, giving higher weights to the faulty classes. In addition, the dataset is balanced by under - sampling the majority class. - **XAI - guided anomaly localization**: - Use the Layer - wise Relevance Propagation (LRP) technique to pass the relevance of the predicted label back to the input image and generate a heat map to accurately locate the defect position. ### Experimental results - **Object detection stage**: Achieved high - precision insulator detection with an mAP50 score of 0.97. - **Shell detection stage**: Also achieved high - precision shell detection with an mAP50 score of 0.98. - **Classification stage**: Through the re - weighting method, the classification accuracy of damaged shells was significantly improved, especially in the "Broken" category, with a performance improvement of up to 12%. - **Comparison with existing methods**: In terms of the average precision of each category, this method is superior to the existing state - of - the - art methods, especially with significant improvements in the "Broken" and "Flash" categories. In conclusion, this paper addresses key challenges in insulator defect detection through innovative methods, especially in the case of dataset imbalance, providing more accurate and reliable detection and localization capabilities.

XAI-guided Insulator Anomaly Detection for Imbalanced Datasets

An Improved Anomaly Detection Model for Automated Inspection of Power Line Insulators

Detection of Power Line Insulator Defects Using Aerial Images Analyzed With Convolutional Neural Networks

Insulator Abnormal Condition Detection from Small Data Samples

A deep learning approach for insulator instance segmentation and defect detection

Efficient Cross-Modality Insulator Augmentation for Multi-Domain Insulator Defect Detection in UAV Images

Data Augmentation of Insulator Aerial Images for Intelligent Inspection of Electric Transmission Systems

Insulator Anomaly Detection Method Based on Few-Shot Learning

Object detection-based inspection of power line insulators: Incipient fault detection in the low data-regime

A Transfer Learning Architecture to Detect Faulty Insulators in Powerlines

Insulator Defects Detection for Aerial Photography of the Power Grid Using You Only Look Once Algorithm

An method for power lines insulator defect detection with attention feedback and double spatial pyramid

Deep inspection: an electrical distribution pole parts study via deep neural networks

Accurate Glass Insulators Defect Detection in Power Transmission Grids Using Aerial Image Augmentation

Deep learning based insulator fault detection algorithm for power transmission lines

E-VarifocalNet: A Lightweight Model to Detect Insulators and Their Defects under Power Grid Surveillance.

High-Voltage Transmission Line Insulator Defects Detection of Leakage Caps on Aerial Images

Power line insulator defect detection using CNN with dense connectivity and efficient attention mechanism

Defect Insulator Detection Method Based on Deep Learning

Aerial Inspection of High-Voltage Power Lines Using YOLOv8 Real-Time Object Detector

Insulator Detection in Aerial Images for Transmission Line Inspection Using Single Shot Multibox Detector