ID-Det: Insulator Burst Defect Detection from UAV Inspection Imagery of Power Transmission Facilities

Shangzhe Sun, Chi Chen, Bisheng Yang, Zhengfei Yan, Zhiye Wang, Yong He, Shaolong Wu, Liuchun Li, Jing FuDrones 2024 · DOI

TL;DR

This paper introduces ID-Det, a novel framework for automatically detecting and localizing burst defects in power line insulators using UAV imagery. By combining a specialized insulator segmentation network (ISNet) with a corner-based defect detector (IBD), ID-Det achieves high accuracy and precision, outperforming general object detection methods and offering a significant step towards automated power infrastructure inspection.

What Problem Does This Paper Address?

The increasing global demand for electricity necessitates robust and extensive power transmission infrastructure. Insulators are critical components that prevent electrical discharge, but they are vulnerable to defects like "bursts" due to their exposure to the environment. Traditional manual inspections are inefficient, risky, and prone to errors. While UAVs have revolutionized power line inspection by providing aerial imagery, the manual interpretation of this data remains a bottleneck. Detecting small insulator defects, especially in complex backgrounds, is challenging. Existing automated methods often struggle with accurate localization or are not specifically tailored for the unique characteristics of insulators, leading to potential misdetections and missed defects.

Methods

ID-Det is a two-stage framework:

  1. Insulator Segmentation Network (ISNet): This component focuses on accurately segmenting insulators from complex backgrounds.
  2. Insulator Clipping Module (ICM): This novel module is key to ISNet's success. It addresses the challenge of long, narrow insulator shapes with curved edges, which traditional segmentation methods struggle with. ICM uses a PCA module to orient the insulator horizontally and then a Chips Module to slice the insulator into smaller, square "chips." This allows the network to better extract features from each part, leading to improved segmentation of edges.
  3. Backbone, Neck, and Head: ISNet utilizes a ResNet50 backbone with a Feature Pyramid Network (FPN) and a detection head that predicts category, polar centerness, and mask. It employs a polar coordinate representation for mask prediction, simplifying the task and improving efficiency.
  4. Chips Stitching Module (CSM): After segmentation of individual chips, CSM reassembles them into a complete insulator mask.
  5. Insulator Burst Detector (IBD): This component leverages the segmented insulator mask to pinpoint burst defects.
  6. Corner Point Extraction: It uses an improved FAST CPDA algorithm to extract key corner points from the insulator mask edges.
  7. Distance-Based Clustering: Corner points are clustered based on their distance to a main direction line derived from the insulator mask.
  8. Periodic Distribution Feature Discriminator: This module analyzes the distances between neighboring corner points within each cluster. Insulator disks normally exhibit a symmetrical and periodic distribution. Deviations from this pattern, specifically larger distances between adjacent corner points than expected, indicate a burst defect. A threshold of 2.5 times the average distance between neighboring points is used to identify these defects.

Datasets

  • ID Dataset: The authors constructed a new dataset named the "ID Dataset" specifically for this research. It comprises 1614 individual insulator instances captured by UAVs. The dataset is formatted in the COCO format.
  • Training set: 1453 images
    • Validation set: 161 images
    • Defective insulators: 551 images (34.14%)
    • The dataset includes images captured from various angles, distances, and resolutions to ensure applicability.

Key Insights and Ideas

  • The "Chip" Approach for Segmentation: The Insulator Clipping Module (ICM) is a clever way to handle the challenging aspect ratio and complex edges of insulators. By breaking down the insulator into smaller, manageable "chips," the network can better learn local features and produce more accurate segmentations, especially at the edges.
  • Leveraging Geometric Properties for Defect Detection: Instead of relying solely on deep learning for direct defect detection, IBD uses the inherent geometric properties of insulators (symmetry and periodic distribution of disks) combined with corner detection. This approach is more robust for localizing defects precisely.
  • Polar Coordinates for Mask Prediction: Using polar coordinates for mask representation simplifies the segmentation task and improves efficiency, making it more akin to target detection.
  • The Importance of Edge Refinement: The paper highlights that while general segmentation methods can identify insulators, refining the edges is crucial for accurate defect localization, which ID-Det achieves.

Key Results

  • Overall Performance (ID Dataset):
    • Accuracy: 97.27%
    • Precision: 97.38%
    • Recall: 94.56%
  • ISNet Performance: Achieved a 27.2% improvement in Average Precision (AP) for segmentation compared to the baseline (PolarMask without ICM).
  • Comparison with General Object Detectors: ID-Det outperformed methods like YOLOv5, YOLOv7, YOLOv8, YOLOv9, and Swin Transformer in accuracy, precision, and recall. For example, it showed a 4.33% increase in accuracy, 5.26% increase in precision, and 2.364% increase in recall compared to general object detection methods.
  • Ablation Studies:
    • The ICM module significantly improved segmentation performance.
  • Baseline + PCA Module and Baseline + Chips Module showed over 10% mAP improvement in segmentation.
  • ID-Det (Baseline + PCA + Chips) achieved a 27.2% mAP improvement over the baseline.
  • For bounding box prediction, ID-Det showed a 5.3% increase in AP compared to the baseline.

Limitations and Future Work

  • Weather Conditions: The current study did not validate ID-Det on UAV imagery captured under extreme weather conditions (rain, fog, snow). Performance might be affected by low visibility.
  • Image Quality: The clarity of UAV images is crucial. Issues like excessive brightness, inconsistent scale, low resolution, and obscured insulators can lead to false positives and negatives.
  • Shooting Conditions: The method has certain requirements for shooting conditions, such as positioning the insulator centrally and avoiding excessively steep angles.
  • Future Work:
    • Explore the use of multiple sensors (near-infrared, thermal infrared, radar) for more comprehensive monitoring, especially in adverse weather.
  • Investigate image super-resolution techniques to enhance image clarity.
  • Further develop multi-sensor data fusion to improve robustness and generalization.
  • Address false positives and negatives by exploring these advanced techniques.

Summary updated

Abstract

The global rise in electricity demand necessitates extensive transmission infrastructure, where insulators play a critical role in ensuring the safe operation of power transmission systems. However, insulators are susceptible to burst defects, which can compromise system safety. To address this issue, we propose an insulator defect detection framework, ID-Det, which comprises two main components, i.e., the Insulator Segmentation Network (ISNet) and the Insulator Burst Detector (IBD). (1) ISNet incorporates a novel Insulator Clipping Module (ICM), enhancing insulator segmentation performance. (2) IBD leverages corner extraction methods and the periodic distribution characteristics of corners, facilitating the extraction of key corners on the insulator mask and accurate localization of burst defects. Additionally, we construct an Insulator Defect Dataset (ID Dataset) consisting of 1614 insulator images. Experiments on this dataset demonstrate that ID-Det achieves an accuracy of 97.38%, a precision of 97.38%, and a recall rate of 94.56%, outperforming general defect detection methods with a 4.33% increase in accuracy, a 5.26% increase in precision, and a 2.364% increase in recall. ISNet also shows a 27.2% improvement in Average Precision (AP) compared to the baseline. These results indicate that ID-Det has significant potential for practical application in power inspection.
BibTeX
@inproceedings{sun2024id,
  author = {Shangzhe Sun and Chi Chen and Bisheng Yang and Zhengfei Yan and Zhiye Wang and Yong He and Shaolong Wu and Liuchun Li and Jing Fu},
  title = {{ID-Det: Insulator Burst Defect Detection from UAV Inspection Imagery of Power Transmission Facilities}},
  booktitle = {Drones},
  year = {2024},
  doi = {10.3390/drones8070299},
}
Figure