Classification of Plant Leaf Disease Using Deep Learning: Fid-Bau Portal

Classification of Plant Leaf Disease Using Deep Learning

Indira, K. / Mallika, H.

Agriculture is critical in human existence. Practically, 60% of the population is occupied with some sort of farming, either straight forwardly or in a roundabout way. The important factor which determines the quality of crops grown in fields is early detection, classification and severity of plant leaf diseases. Traditional plant disease detection was based on manual design of features and classifiers. The traditional methods faced many challenges in real complex environment such as low contrast, noise in the image being captured. In recent years, deep learning model widely used in image classification task is convolutional neural network, and hence, plant disease detection based on deep learning has important academic research value. In this paper, a CNN model with different convolution layers, AlexNet and MobileNet, was designed to classify the plant leaf diseases of pepper bell, tomato, potato, rice, apple and sorghum with a total of 26 classes. The dataset for pepper bell, tomato, potato, rice and apple is taken from kaggle, whereas sorghum dataset is generated by taking pictures from the field. Along with identification of disease, severity detection is performed on two selected diseases of tomato crop using the best model. Using an available dataset of 24,156 images of diseased and healthy plant leaves, the performance of the model was evaluated. Simulation results revealed an accuracy of 84.24% with CNN model consisting of five layers and an accuracy of 91.19% with pretrained AlexNet model and an accuracy of 97.33% with MobileNet with a learning rate of 0.001. Severity detection was performed on tomato early blight and tomato bacterial spot with levels 1–5 using MobileNet. Simulation result revealed an accuracy of 87.08% and 88.75% for tomato early blight and tomato bacterial spot, respectively.