摘要
目的:针对内印茧分选工作劳动强度大、结果易受主观因素影响的问题,提出一种基于透射图像的内印茧识别方法,并构建出内印茧识别模型。方法:设计了一种能反映蚕茧内部特征的蚕茧透射图像采集装置,通过对透射图像进行滤波、GrabCut等预处理,实现背景分割,然后提取图像颜色特征和纹理特征共10个参量,将其作为网络的输入特征向量,分别训练BP和Adam-BP神经网络模型。结果:Adam-BP神经网络模型的准确率为94%,高于传统BP神经网络,且该模型具有良好的泛化性能。结论:本研究的内印茧识别方法较传统方法更为客观准确,且成功减少了识别过程人工的参与,降低了劳动强度,为实现内印茧分选自动化提供良好基础。
Aims:To solve the labor-intensive problem and that of the result being easily affected by subjective factors of inside-stained cocoon sorting,a method of inside-stained cocoon identification based on transmission images was proposed by the construction of an identification model.Methods:A cocoon transmission image acquisition device was designed to reflect the internal characteristics of cocoons.The background segmentation was achieved by pre-processing the transmission image with filtering and Grab Cut;and then a total of 10 covariates of image color features and texture features were extracted and used as the input feature vectors of the network to train the BP neural network and Adam-BP neural network models,respectively.Results:The accuracy of the Adam-BP neural network model was 94%higher than that of the traditional BP neural network;and the model had good generalization performance.Conclusions:Compared with traditional methods,the results of this method are more objective and accurate.It successfully reduces the manual participation in the recognition process,reduces the labor intensity,and provides a good basis for realizing the automation of cocoon sorting.
作者
涂春梅
孙卫红
邵铁锋
梁曼
TU Chunmei;SUN Weihong;SHAO Tiefeng;LIANG Man(Cocoon and Silk Quality Inspection Technology Institute,College of Mechanical and Electrical Engieering,China Jiliang University,Hangzhou 310018,China)
出处
《中国计量大学学报》
2023年第2期303-310,共8页
Journal of China University of Metrology
基金
浙江省基础公益研究计划项目(No.LGG20E050014)。
关键词
内印茧
透射图像
颜色特征
纹理特征
神经网络
inside-stained cocoon
transmission image
color feature
texture feature
neural network