Tomato Plant Disease Recognition Using EfficientDet

Tomato plant diseases have a considerable effect on agricultural yield, necessitating automated and precise disease detection systems. Conventional diagnostic techniques are time-consuming and require professional intervention, making deep-learning-based solutions a necessity. However, current models are plagued by computational inefficiencies and poor generalization over varying environmental conditions, making them less effective in real-world applications. In order to get around these problems, this research suggests using an EfficientNet-based tomato disease classification system along with a Histogram of Oriented Gradients (HOG) to pull out features. The methodology adopted includes preprocessing steps such as image resizing, normalization, noise removal through Gaussian blur, and data augmentation to make the model more robust. HOG was utilized to extract essential texture and edge-based features and extract disease-specific patterns from tomato leaves. The features that were extracted are then put into groups with EfficientNet, which uses compound scaling to make the network as deep, wide, and clear as possible so that the classification works well and quickly. The model was trained on the Plant Village dataset using transfer learning to enhance generalization. The results of the experiment showed that it was 99.41% accurate, 98.08% precise, 97.37% recall, and 97.02% F1-score, which was better than DenseNet, AlexNet, IBiFPN, and CNN-YOLOv7, in that order. These results show that the model we made is a reliable and efficient way to find diseases in real time, which can help us act faster and do a better job of managing crop health.