EfficientNet-Based Architecture for Tomato Leaf Disease Prediction Using Transfer Learning in Precision Agriculture

Tomato cultivation is a significant field crop operation in the world and a primary source of yield. Tomato plants are more susceptible to many diseases, which impact yield and product quality. The detection of disease by conventional methods, such as visual inspection, takes a long time and is unreliable, thus resulting in a delay in management. This study proposes a new method for disease detection of tomato leaves using EfficientNet, a machine learning algorithm that will ensure accurate results without significant computational slowdowns. It uses the PlantVillage dataset, containing images of tomato leaves. Employing a mix of transfer learning and data augmentation, a specialized Convolutional Neural Network model - in this instance, EfficientNet - has been created. The main aim of the model is to successfully identify major diseases like Early Blight, Late Blight, and Septoria Leaf Spot and provide farmers worldwide with a reliable automated system for early diagnosis and treatment. By increasing disease detection, this research reduces crop loss per pesticide applied, thereby achieving more sustainable and precise farming and economic gain for local farmers. The model is an EfficientNet architecture with batch normalization, dropout, and dense layers to ensure optimal feature extraction and sorting into 11 disease classes—the model utilized transfer learning and adaptive optimization algorithms. The model outputs a test accuracy of 98.37% and an F1 Score of 0.9836, indicating that the model is efficient and trustworthy in disease recognition. This performance is significant in agricultural diagnosis. It reduces losses and promotes sustainable agriculture.