Construction And Application Research Of Grape Non-infectious Disease Recognition Model Based On Deep Learning

As one of the four major fruits in the world,grapes are both nutritious and of high economic value.However,in the process of grape growing,non-infectious diseases often appear,which brings great challenges to grape cultivation and industry.Currently,the identification of non-invasive diseases of grapes mainly relies on manual work,which has the problems of slow identification and low recognition rate.Therefore,this paper constructs a grape non-invasive disease identification model based on the improved YOLOXS algorithm,aiming to improve the speed as well as the accuracy of the automatic identification of non-invasive diseases of grapes.The main work of this paper is as follows:(1)Construction of a grape non-infestation disease dataset.Based on the State Key Laboratory of Plant Pest Biology,this paper collected 1484 original images of non-invasive diseases of grapes,and after annotation and pre-processing,a grape noninvasive disease dataset containing 6800 images was constructed.(2)An improved YOLOXS-based grape non-infestation disease identification model was constructed.To reduce the loss of grape non-infestation disease feature information,three FPCUS modules are added to the backbone network of YOLOXS network in this paper to achieve feature fusion between different depths;to mitigate the influence of natural environment on grape non-infestation disease features,a lightweight CBAM attention mechanism module is introduced at the prediction end of YOLOXS to enhance the extraction of key features;meanwhile In order to avoid degradation of the deep network and to make full use of the shallow features,double residual edges are introduced at the prediction side,and the experimental results compared with the relevant authoritative literature show that the highest recognition accuracy of 98.30% is achieved after the improvement.(3)Development of an applet for non-invasive disease identification of grapes.In this paper,a We Chat applet was developed based on the constructed model for identifying non-invasive diseases of grapes.The applet can identify 17 non-invasive diseases of grapes online and also provides information on morphological characteristics,causes of disease onset,and control suggestions of non-invasive diseases of grapes.It solves the problems of low accuracy and slow speed in identifying non-invasive diseases of grapes by human eyes,and has practicality and extension value.This study shows that the constructed model is not only fast and high in the identification of non-invasive diseases of grapes,but also can cover most of the common non-invasive diseases of grapes,which fills the gap in the image identification of non-invasive diseases of grapes under natural conditions in China and helps grape growers to accurately identify non-invasive diseases of grapes and reduce unnecessary economic losses.This study is an important guide for identifying and controlling non-invasive diseases of grapes.