目标检测
计算机科学
算法
残余物
人工智能
特征(语言学)
最小边界框
计算
特征提取
传输(电信)
干扰(通信)
计算机视觉
级联
电力传输
输电线路
对象(语法)
模式识别(心理学)
Viola–Jones对象检测框架
直线(几何图形)
跳跃式监视
功能(生物学)
变量(数学)
算法设计
传感器融合
图像(数学)
作者
Fudi Ge,Yunfei Ding,Xingtao Wu,Yuxin Si,Lina Wang,Dong Ding,Xichao Wang,Hongwei Zhang
标识
DOI:10.1088/2631-8695/ae2c40
摘要
In order to solve the problems of complex background, variable target scale, and frequent false and missed detections in transmission line foreign object detection, an algorithm based on improved RT-DETR is proposed in this paper. The algorithm enhances the feature extraction capability and background interference suppression by introducing a CRMB module with integrated inverted residual shift module (iRMB) and cascade group attention (CGA). In addition, a SSFF-Slimneck cross-scale feature fusion network is proposed to mitigate the information loss during feature fusion. Focaler-Shape-IoU is adopted as the bounding box loss function to accelerate model convergence, enhance generalisation capability and improve detection performance. The experimental results show that the proposed method improves 3.3% and 2.3% on mAP@50 and mAP@50:95, respectively, while the parameters and computation are reduced by 24.5% and 16.4%, respectively. This indicates that the proposed method achieves higher detection accuracy while reducing the computational complexity, which significantly improves the foreign object detection capability of transmission lines.
科研通智能强力驱动
Strongly Powered by AbleSci AI