首页|基于Cache-DCN YOLOX算法的交通标志检测方法研究

基于Cache-DCN YOLOX算法的交通标志检测方法研究

扫码查看
针对传统方式识别交通标志算法存在的检测精度较低的问题,提出了一种基于Cache-DCN YOLOX算法的交通标志识别方法;在该方法中,使用DCN可变形卷积替换backbone中的普通卷积,有效地增大了模型的感受野,提高了特征提取能力;使用EIoU损失函数代替YOLOX中的GIoU损失函数,优化了训练模型,提高了收敛的速度;优化设计了 YOLOX算法中的强弱两阶段的训练过程,增强了模型的泛化性能,同时加入cache方案,进一步提高了检测精度;在交通标志数据集TT100K上进行了实验,提出方法的检测精度为67。2%,比原YOLOX算法的检测精度提升了 6。4%,同时,在被遮挡的小目标等多种受干扰的环境下,提出的方法能够精确地检测出交通标志,并有着较好的置信度,满足实际需求。
Research on Traffic Sign Detection Method Based on Cache-DCN YOLOX Algorithm
Aiming at the problem of low detection accuracy of traditional traffic sign recognition algorithms,a traffic sign recogni-tion method based on cache-convolutional network(Cache-DCN)YOLOX is proposed.In this method,the deformable convolutional network(DCN)is used to replace the ordinary convolution in the backbone,which effectively increases the receptive field of the mod-el and improves the ability of feature extraction;The EIoU loss function is used instead of the GIoU loss function in YOLOX,which optimizes the training model and improves the convergence speed;The training processes of the strength and weak stages in the YOLOX algorithm are optimized and designed to enhance the generalization performance of the model.At the same time,the cache scheme is added to further improve the detection accuracy.Experiments are conducted on the traffic sign dataset TT100K,the detec-tion accuracy of the proposed method reaches by 67.2%,it is 6.4%higher than that of the original YOLOX algorithm,and the pro-posed method can accurately detect traffic signs with good confidence and meet practical needs in various disturbed environments such as occluded small targets.

deep learningYOLOXtraffic signs recognitiondeformable convolutionsmall target detection

高尉峰、王如刚、王媛媛、周锋、郭乃宏

展开 >

盐城工学院 信息工程学院,江苏 盐城 224051

盐城雄鹰精密机械有限公司,江苏盐城 224006

深度学习 YOLOX 交通标志识别 可变形卷积 小目标检测

江苏省研究生实践创新计划项目江苏省研究生实践创新计划项目江苏省高等学校自然科学研究重大项目国家自然科学基金项目江苏省高校自然科学研究面上项目江苏省高校自然科学研究面上项目江苏省自然科学基金项目

SJCX22_1685SJCX21_151719KJA1100026167310818KJD51001019KJB510061BK20181050

2024

计算机测量与控制
中国计算机自动测量与控制技术协会

计算机测量与控制

CSTPCD
影响因子:0.546
ISSN:1671-4598
年,卷(期):2024.32(2)
  • 5