Design And Implementation Of Image Recognition Research Platform For Stored Grain Insect

China is a large grain-storing country.Grains may be damaged by insects during storage.Every year,the country needs to spend a lot of manpower and material resources to ensure the safety of grain storage.In order to protect grains from insects,timely collect information on the occurrence and development of insects in granaries,many automatic monitoring systems for stored grain insects have emerged.In this context,the detection and classification algorithm of stored grain insects based on image recognition technology has been rapidly developed.But there are also problems in the time-consuming and labor-intensive image collection of stored grain insects and the lack of sharing platforms for stored grain insects datasets.Besides,in the management of grain storage,the grain storage custodians and the majority of grain storage farmers learn and search for information on stored grain insects through books and wall charts.The learning efficiency and timeliness are insufficient,and the content update speed is slow.Based on the above background,this paper designed and implemented a research platform for image recognition of stored grain insects through the research and analysis of the field of grain storage industry and image recognition of stored grain insects.The platform provided a channel for uploading images of stored grain insects,and established a sharing platform for the storage grain insect image datasets,and realized a stored grain insect intelligent illustration guide and image recognition APP.The main work done in this paper is as follows:1.Through the study of Java Web,Android,and front-end related technologies,the design scheme of image recognition research platform for stored grain insects was determined with the investigation of the learning status of grain storage personnel and the analyzation of the problems of image recognition of stored grain insects.2.Using Java Web and related technologies and the hierarchical multi-module design idea,this paper developed the background management system,Rest service system,Solr search system,FastDFS file system and login system,which realized the effective management of system resources.This paper used React and Ant Design to design and develop the front-end website for grain storage insect image recognition research,and realized the home module,insect image dataset module,insect illustration module,insect image recognition module and upload insect image module.3.Aiming at the inconvenience of grain storage personnel and grain storage farmers in learning and inquiring information about stored grain insects,a mobile app software was developed using Android technology,which mainly includes illustration guide,image recognition and image uploading functions of stored grain insects.4.Based on the image dataset of stored grain insects,a VGG model was used to train a classification algorithm of stored grain insects,which can identify 10 kinds of stored grain insects in 6 categories with an average accuracy of 0.95.And this paper also realized the recognition of grain storage insects on the server with calling the algorithm through the Java interfaces of TensorFlow.The Resnet model was used to train a stored grain insect image recognition algorithm with less network parameters.And it was successfully ported to the Android mobile phone through the interfaces of TensorFlow Lite,which realized the local identification of the stored grain insect image on the Android mobile phone.5.This paper tested the use cases of each functional module of the system to optimize performance and achieve the expected results.