输电线路巡检中改进YOLOv10的缺陷检测算法研究

Transmission Line Inspection Algorithm Based on Improved YOLOv10

摘要:

针对无人机进行电力巡检时对于输电线路缺陷的检测精度较低的问题，提出了一种基于YOLOv10n的改进缺陷检测算法。其具体结构为在Backbone中添加轻量级卷积神经网络注意力模块CBAM，使得改进后的网络模型在通道和空间两个方面更加关注输电线路导线缺陷的特征，降低漏检、错检率。将YOLOv10n中原来的特征融合框架替换为双向特征金字塔网络BiFPN，该网络在原始的FPN模块中添加了上下文信息的边，并对每个边乘以一个相应的权重，通过不同的权重映射不同的学习特征，因此增加对贡献较大的特征映射。在空间金字塔池化模块通过结合ELAN，使模型能更有效地识别小目标特征。经过一系列的实验证明，改进后的模型准确率达到85.8%，召回率达到80.8%，mAP达到87%，各种指标在一定程度上都得到了提升。由此可见改进的算法提高了检测精度，在输电线路巡检中具有较广泛的应用前景。

Abstract:

An improved defect detection algorithm based on YOLOv10n is proposed to address the issue of low detection accuracy for transmission line defects during UAV power inspections. The specific structure involves adding a lightweight Convolutional Neural Network (CNN) attention module, CBAM, to the Backbone, enabling the enhanced network model to focus more on the features of transmission line conductor defects in both the channel and spatial dimensions. This modification reduces the rates of missed and false detections. Additionally, the original feature fusion framework in YOLOv10n is replaced with a Bidirectional Feature Pyramid Network (BiFPN). This network adds context information edges to the original FPN module and multiplies each edge by a corresponding weight. By assigning different weights to various learning features, the network emphasizes the feature mappings with greater contributions. The integration of ELAN in the spatial pyramid pooling module further enhances the model’s ability to detect small target features. Experimental results show that the improved model achieves an accuracy of 85.8%, a recall rate of 80.8%, and a mAP of 87%. These indicators demonstrate significant improvement, indicating that the proposed algorithm enhances detection accuracy and has broad application potential in transmission line inspection.