Early Detection of Crop Disease With Automatic Image Based Classification Using CNN and Trans-fer Learning

In the current machine vision technology, accurate detection and classification of the crop dis-eases can protect against spoilage. Different diseases of tomato leaf have similar features or traits, making image disease detection confusing and challenging. Farmers cannot recognize whether a crop is infected or not just by looking at its leaves, because the healthy and infected crops resemble the same at first. Deep learning models can be used to overcome this prob-lem within less computational time. As a result, a new framework is implemented in this work through fine tuning the Deep Convolutional Neural Networks (DCNN) model using hyper parameters like learning rate, batch size, and epochs by applying transfer learning techniques for detecting tomato leaf disease. The data in this work is collected from the Plant Vil-lage database, which includes 20,639 images. The pro-posed model is implemented on three pre trained DCNN models-Alex Net, ResNet50 and VGG16. The proposed framework attains highest classification ac-curacy of 99.26% for fine tuning DCNN. The simula-tion results demonstrates that the fine-tuning Res-Net50 performs better classification of crop diseases when compared to the other DCNN models.