A Hybrid Deep Learning Based Approach For Tomato Disease Classification In Natural Environment

Abstract

Tomatoes are widely cultivated across the globe, boasting a rich array of nutrients like vitamin C and a delicious taste, making them essential for agricultural production and widespread consumption. The threat of plant diseases poses a significant challenge to global agriculture. The early detection and accurate identification of plant diseases are crucial for preventing crop losses and minimizing the use of harmful pesticides. Deep learning models, with their ability to extract complex features from images, have revolutionized the field of computer vision, including plant disease classification. The available CNN architectures like efficientNet, MobileNet etc. proven the remarkable results in various domains, but other than CNN architectures, dataset also plays the vital role for any problem solved using deep learning, similarly for Plant Disease detection. PlantVillage is the most widely used publicly available dataset but prepared in controlled environment, other available dataset is PlantDoc having images near to real-world. But earlier methods not able to provide the good accuracy on PlantDoc dataset, therefore We propose a hybrid technique based on a combination of object detection and classification CNN architectures, which achieved 90.7% accuracy as well as resilient F1 score and recall metrics. The suggested methodology also achieves a high level of accuracy of 85.2% using the lightweight CNN architecture MobileNet, allowing the model to be deployed in mobile/embedded devices.