Improved Deep Learning Models for Plants Diseases Detection for Smart Farming

Traditional farming methods consume time and effort, which affects productivity. Smart agriculture aims to improve decision-making and crop management using IoT's connectivity and data analysis capabilities. Plant diseases result in significant financial losses in the farming sector. Accurately detecting of diseases is crucial for ensuring the long-term sustainability of agriculture. Deep learning, has recently garnered significant attention for plant and weed detection, disease diagnosis, and pest classification in agricultural industries. In this paper, the most previous studies have been discussed focusing on a several plant species and a specific type of disease. The dataset used in these models is (PlantVillage). The dataset includes 14 types of plants images with 39 different classes of plant diseases. We propose an enhanced deep learning models (EfficientNetB2, Xception, ResNet50) by adding a custom classification layer, which significantly improved the model's accuracy and classification performance. The improved models achieved accuracy as: (EfficientNetB2 is 97.70% , ResNet50 is 97.86% and Xception is 98.97%). The main aim of this paper is to enables the farmers to detect plant diseases in early stage of disease without consulting experts.