Research On Key Issues Of Chinese Herb Cultivation Based On Convolutional Neural Network

The quality of Chinese herbs largely depends on how well the Chinese herb crop is cared for during the cultivation of Chinese herbs.This thesis is based on Chinese herb seed images for Chinese herb seed identification,field weed detection by field images,Chinese herb yield prediction using meteorological,soil,and yield data,and the design and implementation of a Chinese herb planting service platform to provide auxiliary decision-making for Chinese herb planting.The thesis work is as follows:1.To address the problems of time-consuming and laborious manual identification of Chinese herb seed and the low accuracy of sampling identification,selective kernel network-based identification of Chinese herb seed is proposed.Since the selective kernel network uses a fully connected network in the operation of dimensionality reduction and then dimensionality increase,which causes too much computation,this thesis uses efficient attention channels to improve the one-dimensional convolution to reduce the computation of the network,and the improved model is called ECA-SKNet.The proposed model achieves 93.04% classification accuracy in maize seeds dataset and 98.56%classification accuracy in castor seeds dataset.2.To address the problems that manual weeding is labor-intensive and indiscriminate spraying of pesticides causes bad effects on the quality of Chinese herbs such as spotting,perforation,scorching and even wilting,a weed detection algorithm based on one-stage objective detection for Chinese herbs in the field is proposed.The backbone network of YOLO v3 is improved by replacing the residual blocks with the cross stage partial structure to reduce the computation of the network,and the improved model is called CSP-YOLO3.The experimental results show that the m AP of the CSP-YOLO3 model is82.5%,which is the best result among all models.Its detection frame rate reaches 31.25 frames per second,which meets the requirement of real-time detection.3.To address the problems that the existing yield prediction methods with lossy prediction will cause damaging operations such as slicing of Chinese herbs and the high cost of using hyperspectral and remote sensing data,a yield prediction model for Chinese herbs based on temporal convolutional network is proposed.The method is based on temporal convolutional network and improved on it to reduce the computational effort,using meteorological data and soil data combined with yield data as training data for yield prediction,and the improved model is called TCN-Lite.The experimental results show that the root mean square error of TCN-Lite for the yield data set is 4.95 and 17.03,and the Pearson correlation coefficient is 74.11% and 85.61%,respectively.4.A service platform for Chinese herbs planting process was designed and implemented.The platform uses a browser/server architecture,programming language using Java language,back-end development framework using Spring Boot framework,front-end UI using Bootstrap,front-end framework using Vue,database using open source database My SQL.The platform includes three parts: user management,planting management and data management,among which planting management provides users with Chinese herb seed identification,field weed detection and Chinese herb yield prediction.