改进Faster R-CNN的输电线路山火图像检测方法  

Improved Faster R⁃CNN method for detecting wildfire images in transmission lines

作　　者：黄力[1] 吴珈承 HUANG Li;WU Jiacheng(College of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443002,China)

机构地区：[1]三峡大学电气与新能源学院,湖北宜昌443002

出　　处：《现代电子技术》2025年第9期173-179,共7页Modern Electronics Technique

基　　金：湖北省自然科学基金项目(2020CFB376)。

摘　　要：针对山火严重威胁输电线路安全的问题,提出一种改进Faster R-CNN的输电线路山火图像检测方法。选用ResNeSt50作为主干网络以提升模型性能,同时在主干网络后面加入递归特征金字塔(RFP)以增强模型在多尺度上的特征提取能力。采用CIoU Loss回归损失函数以提高边界框回归速率和定位精度,使用Focal Loss分类损失函数以提高对小目标的烟雾和火焰检测精度。运用Kmeans++聚类算法对烟雾和火焰数据进行anchor尺寸优化,以提高算法的检测准确率。利用数据增强技术来解决图像数量不足和天气环境变化影响检测精度的问题。经过训练和测试,结果显示改进后的Faster RCNN方法在平均精度均值上达到了95.54%,比原模型提高了7.39%,能够有效识别输电线路附近产生的烟雾和火焰,满足山火检测准确性和实时性的要求。In view of the serious threat of wildfires to the safety of transmission lines,an improved Faster R-CNN image detection method for transmission line wildfires is proposed.The ResNeSt50 is selected as the backbone network to improve model performance,and the recursive feature pyramid(RFP)is added behind the backbone network to enhance the model′s feature extraction ability at multiple scales.The CIoU Loss regression loss function is adopted to improve the bounding box regression rate and localization accuracy,and the Focal Loss classification loss function is used to improve the accuracy of smoke and flame detection for small objects.The Kmeans++clustering algorithm is used to optimize anchor size for smoke and flame data,so as to improve the detection accuracy of the algorithm.The data enhancement technology is used to eliminate the facts that insufficient images and weather environment changes will affect the detection accuracy.After training and testing,the results show that the improved Faster R-CNN method achieves a mean average precision(mAP)of 95.54%,which is 7.39%higher than that of the original model.To sum up,it can effectively identify smoke and flames generated near transmission lines,meeting the requirements of accuracy and real-time detection of wildfires.

关 键 词：深度学习 山火检测 烟雾检测 Kmeans++ ResNeSt50 CIoU Loss Focal Loss RFP 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TM75[电子电信—信息与通信工程] TP391.41[电气工程—电力系统及自动化]