摘要
在药片残缺检测分类任务中,为衡量分类模型的精度与模型大小,保证模型精度的同时,减少计算量和参数量。提出一种改进的GhostNet轻量化神经网络,保证在加工过程中能够有效地检测出残缺药片。构建数据集并增强处理,压缩网络模型,将注意力机制升级为ECA模块。实验结果显示,改进后的网络模型能够达到98.85%的分类精度,模型参数量仅为0.635×10^(6)。与其它轻量化网络进行了对比分析,取得了更高的分类精度,更少的参数量和计算量,更优的分类性能。
In the task of tablet defect detection and classification,in order to measure the accuracy and size of the classification model,ensure the accuracy of the model,and reduce the amount of calculation and parameters.In this paper,an improved GhostNet lightweight neural network is proposed to ensure that incomplete tablets can be effectively detected in the processing process.Build data sets and enhance processing,compress network models,add dropout layer to prevent over fitting,and upgrade attention mechanism to ECA module.The experimental results show that the improved network model can achieve 98.85%classification accuracy,and the model parameter is only 0.635×10^(6).Compared with other lightweight networks,it achieves higher classification accuracy,less parameters and computation,and better classification performance.
作者
黄开坤
徐兴
Huang Kaikun;Xu Xing(School of Mechanical Engineering,University of South China,Hengyang 421001)
出处
《现代计算机》
2022年第19期81-86,共6页
Modern Computer
