Analytical Study of Deep Learning Architectures for Classification of Plant Diseases

Abstract

Plant disease has a significant impact on crop yield of agriculture resulting in increased economic losses. Plant disease detection is a real hurdle in the agricultural area. Farmers have difficulties is identifying, diagnosing and classifying the actual cause of diseases with the naked eye. Furthermore, identifying the diseases takes time and requires a well-trained person and specialist. In larger farms, disease detection becomes more difficult and challenging task. To address these issues, an automatic computerized plant disease identification and classification system based on Image Processing and Deep Learning techniques could be used to recognize diseases at an earlier stage as well as increase the yield. The Deep Learning technique is more accurate, efficient and reliable than the Machine Learning technique in detecting plant diseases. The goal of this analytical study is to evaluate and compare the accuracy of Deep Learning architectures AlexNet, GoogLeNetand DenseNetwith different optimizers viz SGD, RMS, PROP for tomato plant disease identification and classification.The results showthat the GoogLeNet in combination with ADAM optimizer performed well with an an accuracy of 99.56 % in identification and classification of tomato plant disease. Further, The GoogLeNet architecture trained well with the Adam optimizer achieved the best F1-score of 99.53% as compared to AlexNet and DenseNet with other optimizers.  The proposed model GoogLeNet with ADAM optimizer is beneficial for farmers in identifying and classifying tomato diseases as it has high success rate.