摘要
棉纺产线中的自动化、数字化设备提高了产线的生产效率与质量,但在当前仍存在棉纺产线管道棉流中棉花状况难检测的问题。因此,结合棉纺产线中棉流图像的特征,提出了一种基于轻量级网络的分割方法(Light cotton-net)。网络基于一种对称编解码结构,通过优化卷积方式与上采样方法、设计特称提取结构,在保证分割精度在误差可接受范围内的同时大幅减少网络参数、提高网络预测速度。以异纤机中拍摄的棉流图像为数据集,加入随机偏移、缩放、亮度变换等数据增广操作。实验数据表明,在网络参数量6.0M(million),预测每张图片时间为35.328ms的情况下,模型的分精确度和召回率分别为96.63%和93.87%,模型分割精度基本与U-net网络等同,参数量约为其1/3,图像分割速度约为其5倍,模型对系统内存及算力的需求更低,更适合在工业设备上的部署。
The automation and digital equipment in the cotton production line have improved the production efficiency and quality of the production line,but there is still the problem that the cotton condition in the pipe flow of the cotton production line is difficult to detect.Therefore,based on the characteristics of cotton flow images in cotton spinning lines,this paper proposes a segmentation method based on lightweight network(Light cotton-net).The network is based on a symmetric encoding and decoding structure.By optimizing the convolution method and upsampling method and designing the special name extraction structure,the network parameters are greatly reduced and the prediction speed is improved while the segmentation accuracy is within the acceptable range of error.The data set of cotton stream images taken in the fiber machine was used to add random migration,scaling,brightness transformation and other data augmentation operations.Experimental data show that the segmentation accuracy and recall rate of the model are 96.63%and 93.87%respectively when the number of network parameters is 6.0m(million)and the prediction time of each image is 35.328ms.The segmentation accuracy of the model is basically the same as that of the U-NET network,the number of parameters is about 1/3,and the image segmentation speed is about 5 times.The model requires less memory and less computing power and is more suitable for deployment on industrial equipment.
作者
李尤
张晨
魏巍
向森
LI You;ZHANG Chen;WEI Wei;XIANG Sen(College of Information Science and Engineering,Wuhan University of Science and Technology,Hubei Wuhan 430081,China;Wuhan Zhimu Intelligent Technology Partnership,Hubei Wuhan 430074,China)
出处
《计算机仿真》
2024年第1期395-400,共6页
Computer Simulation
基金
国家自然科学基金资助项目(61702384)。
关键词
棉花图像
图像分割算法
对称结构网络
棉流数据集
Cotton image
Image segmentation algorithm
Symmetric structure network
Cotton stream data set
