Real-time Production Monitoring for Chinese Pharmacy Based on Convolutional Neural Network

摘要

A deep monitoring architecture for extraction and esterification process monitoring in Chinese Pharmacy production based on convolution neural network is proposed and lightweight model is also researched. Different convolution kernel including 2D convolution, depth-wise convolution and 3D convolution is compared to improve the performance of monitoring. Experimental results show that accuracy of the designed extraction monitoring model is 98.95% using 3D convolution and esterification monitoring mode has the accuracy of 98.68%. Meanwhile, lightweight model using depth-wise convolution can reduce computation while ensuring accuracy.