Plant health: Early tomato disease detection using a lightweight CNN model

Tomatoes are grown in large areas worldwide and are a significant crop. Tomato diseases affect the growth of the plants and reduce their yields. Computer vision and deep neural network techniques have been employed for the timely detection and classification of these diseases. Traditional deep learning algorithms need several parameters and high computation. A lightweight deep neural network model significantly reduces the number of parameters. Light-weighting techniques like point-by-point convolution, deep separable convolution, bottleneck, Inverse residual, and model compression techniques to reduce the number of model parameters. Deep learning architectures like ResNet50, Xception, and MobileNet were pre-trained using ImageNet dataset and further fine-tuned using the tomato leaf dataset. Accuracy of 94% with a parameter size of 14MB is attained by MobileNetV2 model. We realize an early recognition system using mobile for tomato disease leaf images.