Deep learning technique for plant disease detection

Abstract

A nation's economy is primarily reliant on agricultural growth. However, several plant diseases seriously impair crop growth, both in terms of quantity and quality. Due to a lack of subject matter specialists and low contrast data, accurate diagnosis of many diseases by hand is highly difficult and time-consuming. The farm management system is therefore looking for a method for automatically detecting early illnesses. To overcome these challenges and correctly classify the different diseases, an efficient and small deep learning-based framework (E-GreenNet) is proposed. A MobileNetV3Small model is used as the foundation of our end-to-end architecture to produce finely tuned, discriminative, and noticeable features. Furthermore, the new plant composite (PC), plantvillage (PV), and data repository of leaf images (DRLI) datasets are used to independently train our proposed model, and test samples are used to evaluate its actual performance. The suggested model achieved accuracy rates of 1.00 percent, 0.96 percent, and 0.99 percent on the given datasets after a rigorous experimental study. Additionally, a comparative investigation of our proposed technique against the state-of-the-art (SOTA) reveals extremely high discriminative scores.