Research On Vegetable Disease Recognition Technology Based On Machine Vision

Vegetables,rich in vitamins and organic plant protein,function as an essential part in the diet structure of Chinese people,and thus play a particularly important role in the national economy.However,due to the unfavorable planting environment,the frequently-occurring diseases lead to the decline of vegetable yield and quality.In recent years,the agricultural science and technology develops more and more rapidly in China,and many researches have been carried out to explore high-tech ways to solve traditional agricultural problems.Among them,the prevention and control of agricultural diseases has become one of the research hotspots.How the modern science and technology can accurately identify agricultural diseases and provide timely remedies to agricultural producers is the urgent problem to solve.Therefore,by taking the typical early epidemic,late epidemic and leaf mould disease of tomato leaves as samples,this research has a set of vegetable disease monitoring system based on machine vision technology.There are four main points in this paper:(1)The acquired tomato images are preprocessed by scale transformation,denoising and image enhancement,highlighting the distinguishing feature of the disease images.The image background and is also removed so as to reduce the leaves size in the images,screening the processed leaves images for health,and coarse screening for the healthy state of the planting environment.(2)This research has established a vegetable disease recognition model based on the improved MobileNet-V2 network.A research on leaf species identification through neural network reveals that the current identification of mainstream diseases mainly relies on the combination of image processing and neural network.But it is not suitable to run in mobile devices because of its large amount of calculation.In this research,the model is compressed in multiple ways to design a lightweight network structure that can be operated in mobile devices with limited computing power.The result is significant.Compared with the same type of lightweight network,this model gains 30% in the recognition rate and meanwhile loses only 1% in the accuracy rate.(3)It is important to recognize the diseases and foresee their occurrence to a certain extent.Therefore,this study has designed an environmental parameters prediction model based on PSO-SVR.Based on the analysis and training of environmental parameters of tomato growth,the environmental parameters prediction model is established.According to forecast data,combined with the range of tomato infection,the model can provide different degrees of warnings for users as a way to achieve the goal of early warning and early control.The experimental results show that the accuracy rate of this model is 97.6% in temperature prediction and 96.8% in humidity prediction,which can be used as a reference index for early warning.(4)Based on the previous analysis and design of disease identification and early warning,the vegetable disease monitoring system was established,which has the functions of user login,shooting,health screening,disease recognition,environmental prediction and early warning on diseases.After repeated testing,all the functions could work smoothly.At the same time,in order to improve the user experience,the UI is designed in a flat style to simplify the operation interface without compromising on the functions.