Image Based Deep Learning Approach for Plant Disease Detection

Abstract

Agricultural productivity and quality are seriously influenced by plant illnesses, causes endangering global food security. Consequently, early detection and treatment of these diseases should be embraced to mitigate losses while enabling sustainable agriculture. Over the years there has been a great progress due to emergence of deep learning techniques for optimizing the image-based process of detecting plant diseases. The objective of this research is to diagnose agricultural diseases accurately and effectively based on an image-based deep learning approach for plant disease identification. As a suggestion, the method involves using convolutional neural networks (CNNs) to identify appropriate features in plant images that can subsequently be used to determine whether they are healthy or sick. A set of images which include both healthy and diseased plants is employed during training and evaluation processes. The model architecture consists of multiple convolutional and pooling layers to extract relevant features from input images. To prevent overfitting, dropout layers are added, and the model is trained with a small learning rate of 0.0001. The CNN is trained on a dataset of 70,295 training images and validated on 17,572 validation images belonging to 38 different classes of plant diseases. The model achieves a high training accuracy of 97.82% and a validation accuracy of 94.59%. Additionally, the evaluation of a model's performance involves several metrics, including precision, recall, and the F1-score showing promising results for practical application in agriculture.