基于改进Faster R-CNN的小目标电缆线号定位模型  

A Small Target Cable Number Localization Model Based on Improved Faster R-CNN

作　　者：韩境和 于正林[1] HAN Jinghe;YU Zhenglin(School of Mechatronic Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学机电工程学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2023年第1期65-72,共8页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省科技厅基础研究项目(202002044JC)。

摘　　要：针对目前小目标电缆线号难以准确定位的问题,提出了基于深度学习的在不同干扰下能准确定位小目标线号的算法。由于电缆应用场景存在有噪声、粉尘、光照变化等方面的恶劣条件,不利于线号区域的定位,而且小目标线号对定位精度要求较高,因此基于Faster R-CNN模型进行改进,主干网络使用ResNet-50并对其进行优化,利用特征金字塔和多头自注意力机制,提升网络性能,提高小目标线号的检测精度与模型鲁棒性。实验结果表明,提出的电缆线号定位模型算法相比优化之前准确率提升了3.9%,定位准确率高达99.2%,能有效提高小目标线号的定位准确率。For the current problem that small target cable line numbers are difficult to locate accurately,an algorithm based on deep learning that can accurately locate small target line numbers under different disturbances is proposed.Since the cable application scenario has harsh conditions in terms of noise,dust,and light changes,which are not conducive to the localization of the line number area,and the small target line number requires high localization accuracy,the Faster R-CNN model is improved based on the Faster R-CNN model,the backbone network uses ResNet-50 and optimizes it,and the feature pyramid and multi-headed self-attention mechanism are used to improve the network performance and the small detection accuracy and model robustness of the target line number.The experimental results show that the proposed cable line number localization model algorithm improves the accuracy rate by 3.9% compared with that before optimization,and the localization accuracy rate is as high as 99.2%,which can effectively improve the localization accuracy of small target line numbers.

关 键 词：小目标线号定位 深度学习 特征金字塔 Faster R-CNN ResNet-50 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]