Computer-aided Diagnosis Of Plant Disease Based On PD~2SE-Net And L-CSMS

In modern farming,plant disease diagnosis and severity estimation play a critical role in agriculture field,which are of great significance to prevent the spread of diseases,substantially ensure the reduction of economic losses and maintain the sustainable development of agricultural production.Nowadays,there are many plant disease diagnosis works have been presented by using traditional machine vision,the computer vision and image processing has been widely used to address the problem of disease diagnosis.Although the existence of techniques for the detection and diagnosis of plant diseases are valid,they are incapable of solving the problem of plant disease severity estimation.In this work,we proposed a convolution neural network(CNN)-based method to achieve the automatic identification and diagnosis of plant disease.This thesis gives an insight into an automatic identification and severity estimation system to diagnosis the plant disease,a plant disease diagnosis system named PD~2SE-Net based on multi-task combined with the Res Net and Shuffle Net was designed.And the first plant disease detection and severity estimation system integrating plant species classification,plant disease classification and disease severity estimation was developed,which realizes the comprehensive diagnosis of plant disease.Additionally,inspired by Inception-based module and Shuffle Net V2 unit,we proposed a light-weight L-CSMS network to achieve plant disease severity estimation.Extensive experiments have been conducted in this work,the proposed method has achieved excellent performances(overall accuracies of 0.99,0.98 and 0.91 for plant species recognition,plant disease classification and disease severity estimation,respectively)over the existing approaches.In addition,the proposed light-weight models achieved an accuracy of almost 91%with fewer parameters and computation complexity.The main contribution of this work and contributions are as follows:(1)The study of plant disease diagnosis plays an important role in the production of agriculture,few researches have involved in the recognition of disease severity estimation.This thesis put forward a computer-assisted plant disease diagnosis system by using PD~2SE-Net,which is a wonderful system to achieve plant disease diagnosis by employing three branches to identify plant species,disease and disease severity recognition.(2)A light-weight model named L-CSMS was proposed in this work,which consists of residual learning,channel-shuffle operation,and multiple-size convolutional modules.The proposed light-weight CNN-based model achieved a competitive performance over the previous works with fewer parameters.The excellent results for the identification of plant disease severity illustrate the feasibility and effectiveness of the proposed model.(3)The user-operated software based on Windows system was designed for the plant disease diagnosis in this work.The proposed methods including plant disease diagnosis system based on PD~2SE-Net and L-CSMS are embedded in the utility software.