Combined Mobilenet for Plant Diseases Classification

Abstract

Plant disease is a major threat to global food security, as it can reduce crop yields and even lead to complete crop failure. The consequences can be devastating, causing food shortages, price increases, and potentially leading to famine and starvation. Various factors can cause plant diseases, such as fungi, viruses, and bacteria. Each disease has specific symptoms that can be identified by experts, which is crucial for accurate diagnosis and effective disease management. In this work, we propose a new lightweight architecture for classifying plant disease images. This architecture utilizes a pretrained MobilenetV3Large model and a modified version of the fused MBC block to improve the accuracy of plant disease detection. We employ transfer learning and a specialized training methodology to achieve faster convergence and higher accuracy. To evaluate and test our model, we use a dataset containing 7,586 images of pepper, potato, and strawberry diseases. Our results demonstrate that our model outperforms state-of-the-art architectures, achieving a remarkable 99.93% accuracy during the test phase.