摘要
深度学习网络模型参数量大,不适用于嵌入式或移动设备上。针对工业钢材生产过程中的实时检测问题,提出了一种改进的R-Tiny-YOLOv3工业钢材瑕疵检测算法。首先,在Tiny-YOLOv3结构中加入残差网络结构,提高检测的精度。增加了空间金字塔SPP网络模块,提高网络特征提取能力。结合不同网络层的特征信息,将检测提高到三个尺度。然后,选取CIOU作为损失函数,使目标检测框的回归更加稳定。最后对数据集进行数据增强,并在Cambricon 1H8嵌入式平台进行测试。实验结果表明改进的R-Tiny-YOLOv3算法能够实时地检测出瑕疵目标,平均准确率提高了10.8%,运算速度可达39.8帧/s,为工业钢材瑕疵检测的嵌入式应用提供了参考。
The deep learning network model has a large number of parameters and is not suitable for embedded or mobile devices.Aiming at the problem of real-time detection in industrial steel production,an improved R-Tiny-YOLOV3 defect detection algo-rithm for industrial steel was proposed.Firstly,the residual network structure is added into the Tiny-YOLOv3 structure to im-prove the detection accuracy.The space pyramid SPP network module is added to improve the ability of network feature extraction.Combined with the characteristic information of different network layers,the detection can be improved to three scales.Then,CIOU is selected as the loss function to make the regression of the target detection box more stable.Finally,the data set was en-hanced and tested on Cambricon 1H8 embedded platform.The experimental results show that the improved R-Tiny-YOLOV3 al-gorithm can detect defect targets in real time,the average accuracy is increased by 10.8%,and the operation speed can reach 39.8 frames/s,which provides a reference for embedded application of industrial steel defect detection.
作者
章曙光
刘洋
张文韬
王浩
ZHANG Shu-guang;LIU Yang;ZHANG Wen-tao;WANG Hao(School of Electronics and Information Engineering,Anhui Jianzhu University,Anhui Hefei 230601,China;Information Network Center,Anhui Jianzhu University,Anhui Hefei 230601,China)
出处
《机械设计与制造》
北大核心
2024年第5期97-101,共5页
Machinery Design & Manufacture
基金
赛尔网络下一代互联网创新项目(NGII20190602)
安徽省教育厅自然科学重点项目(KJ2016A155)
赛尔网络下一代互联网技术创新项目(NGII20170117)。