Multiscale Based MobileNet V3 Apple Leaf Disease Identification And Application

In recent years,the demand for apples has been increasing year by year,leading to an increase in apple planting area and yield.However,large-scale planting and high yield often lead to the spread of apple diseases,resulting in significant production and economic losses for fruit farmers.Therefore,timely and accurate identification of apple leaf diseases is of great significance.The traditional identification of apple leaf diseases mainly relies on the experience of fruit farmers to determine which disease the apple tree is infected with.This method is not only time-consuming and labor-intensive,but also prone to misdiagnosis or missed diagnosis.However,using machine learning methods for feature extraction of apple leaf diseases is very difficult,and traditional feature extraction methods cannot extract key features,resulting in decreased recognition accuracy and other issues.In order to solve the above problems,this study used public data sets and web crawler to sort out the data sets including six apple leaf diseases and one healthy apple leaf.Through a series of benchmark experiments,improvement experiments,and result analysis,high-precision identification of apple leaf diseases has been achieved,and corresponding system design has been carried out to further increase the practicality of the research.The main work of this article includes the following four points:(1)Build a dataset of apple leaf diseases.In view of the problem that there are few public data sets of apple leaf diseases based on complex environment,six common apple leaf diseases and one healthy apple leaf image were collected from the open data set of fly pulp and web crawler,respectively: powdery mildew,spotted leaf disease,black spot,mosaic,gray spot,rust,and healthy leaves.And the dataset was expanded to obtain a total of 8332 apple leaf disease datasets.(2)The MobileNet V3 model based on transfer learning is constructed.Benchmark experiments were conducted on Google Net,VGG 16,Res Ne 50,MobileNet V2,Inception V3,and MobileNet V3 models.By considering the accuracy and practicality of the models,MobileNet V3 was selected as the benchmark model for subsequent research.In order to further improve the generalization ability of the model,transfer learning was introduced.Compared with the benchmark model,the accuracy of the transferred model was improved by 2.5%.(3)Build a multi-scale based Multi MobileNet V3 model.In response to the problem of uneven lesion area in the apple leaf disease dataset constructed by our research institute,this study analyzed the shortcomings of the original MobileNet V3 and made improvements.Through multi-scale feature fusion,high-level features and low-level features rich in detail information are fused to obtain more and more accurate target information.Compared with the model only using transfer learning,the accuracy of the improved model is improved by about 2%.(4)Design an apple leaf disease identification system.Currently,research on identifying apple leaf diseases mostly remains in the stage of establishing recognition models.This article studies the construction of an apple leaf disease identification system based on the improved model,and ultimately realizes the identification and prevention measures recommendation of apple leaf diseases.Convenient for fruit farmers to accurately diagnose apple leaf diseases and take corresponding prevention and control measures in a timely manner,thereby ensuring the healthy growth of apple trees and apple yield.