摘要
提出一种基于改进Faster R-CNN(region-convolutional neural networks)的车辆识别算法,用于处理不同类别车辆的识别问题.为了解决部分外形相似类别的车辆之间的误检问题,该方法使用空洞卷积来提高感受野,结合空洞空间金字塔池化(atrous spatial pyramid pooling,简称ASPP)来增强多尺度信息的获取,以此来增强网络对外形相似车辆之间差异的敏感性,提升算法的准确率.实验结果表明,改进的Faster R-CNN模型mAP值达到93.45%,具有较高的精确度、较小的误检率和更好的鲁棒性.
This paper proposed a vehicle recognition algorithm based on the improved Faster R-CNN(region-convolutional neural networks)to solve the problem of identifying different types of vehicles.In order to solve the problem of false detection between vehicles with similar appearance,this method used atrous convolution to improve the receptive field,combined with atrous spatial pyramid pooling(ASPP)to enhance the multi-scale information which increased the sensitivity of the network to differences between vehicles of similar appearance and improved the accuracy of the algorithm.The experiment results showed that the mAP value of the improved Faster R-CNN model reached 93.45%,which had higher accuracy,smaller false detection rate and better robustness.
作者
宁俊
王年
朱明
NING Jun;WANG Nian;ZHU Ming(School of Electronics and Information Engineering, Anhui University, Hefei 230601.China)
出处
《安徽大学学报（自然科学版）》
CAS
北大核心
2021年第3期26-33,共8页
Journal of Anhui University(Natural Science Edition)
基金
国家自然科学基金资助项目(61772032,61672032)。
