Revolutionizing Plant Health: Leveraging EfficientNet B3 for Advanced Tomato Disease Detection in Agricultural AI

Addressing the challenge of disease detection in tomato plants, this research introduces a novel approach using deep learning, specifically through the EfficientNet B3 model. Traditional methods, often limited by subjectivity, are surpassed by this model, known for their efficiency and accuracy. Our study employs a dataset of approximately 20,000 images, encompassing various tomato plant diseases and healthy specimens. The EfficientNet B3 model is adeptly adapted to this context, initially trained on raw image data for precise disease classification. A key innovation is the implementation of the model with various types of inputs images to set the best image size to be given to model. We conducted experiments with image resolutions of 64×64, 128×128, 256×256, and 300×300 pixels, finding that 300×300 yields the best results with an accuracy of 94.91%. This fine-tuned model outperforms traditional models like ResNet50 and VGG16, offering a significant increase in the accuracy to rightly predict the class of the disease in tomato plant. The model's effectiveness is assessed using F1 score, recall, accuracy, and precision. This study demonstrates the potential of deep learning in agricultural applications and sets a precedent for future research in plant disease detection, efficiently balancing precision.