Enhanced Plant Stress Classification using Deep Convolutional Neural Networks with MobileNetV2 and Transfer Learning
Pratham Kaushik, Saniya Khurana
Abstract
Plant stress is a state of plants caused by growth conditions that are less than ideal. This includes major effects on plant health, growth, and yield. Stress factors are mainly categorized as abiotic or biotic, relating to light, water, temperature, pathogens, and pests. Proper classification of the type of stress being exhibited by a plant is important, so prompt intervention and mitigation may occur. The paper presents a completely different concept of plant stress classification using convolutional neural networks with MobileNet. This model was trained and tested on a dataset that includes diversity for the classification of stresses into five categories: bacterial leaf blight, blast, healthy, hispa, and leaf spot. The trained model achieved high precision, recall, and f1-scores with an overall accuracy of 89%. These results show the impressive potential of deep learning models in the accurate identification and classification of plant stress, thus helping in the development of some targeted strategies for the management of stress and the protection of crops.