摘要
针对传统神经网络识别算法的结构复杂,参数量大,识别率低,边缘部署困难等问题,本文基于集成管理工具anaconda、开发环境pycharm实现Mobilenet V2训练算法,多比例拆分数据集,优化深度残差模块、批处理参数和dropout参数;改进的MobileNet V2模型在瓜子、花椒、虾、杏仁色选分类任务上的精度均能达到工业标准;最终在边缘部署芯片魂芯五号(HXAI 100)上实现了色选机领域智能分类的效果.
Abstract
Aiming at the problems of complex structure,large number of parameters,low recognition rate,and difficult edge deployment of traditional neural network recognition algorithms,this paper implements the Mobilenet V2 training algorithm based on the integrated management tool anaconda and the development environment pycharm,splits the dataset in multiple proportions,and optimizes the deep residual module,batch parameters and dropout parameters.The improved MobileNet V2 model can meet industry standards in the task of sorting and sorting melon seeds,peppercorns,shrimp,and almonds.Finally,the intelligent classification effect in the field of color sorter was realized on the edge deployment chip Soul Core V(HXAI 100).
NETL
NSTL
万方数据