Classification of Tomato Leaf Diseases using Deep Learning

Abstract

In the rapidly developing field of deep learning, using Hybrid Convolutional Neural Networks (CNN) and Artificial Neural Networks (ANN) improves both the process of creating features and making important decisions. The study looks at using CNNs and ANNs together, mainly investigating how transfer learning can boost their success when the datasets are small.
Since data in deep learning applications is not always evenly distributed, the models may end up with bias and have difficulty being applied in general ways. Many different approaches are tried in this study, including oversampling, undersampling, class-weight balancing and data augmenta- tion, to make sure the outcomes are also strong on datasets with imbalances in the classes.
Django framework handles the backend and MySQL takes care of database management in the proposed system. The user interface is developed with HTML, CSS, JavaScript and Bootstrap in the front end. The findings also show that building special versions of models for energy-efficient use on edge devices using TensorFlow Lite and ONNX is very helpful.
The results of this study offer important facts about setting up hybrid models, keeping them simple enough to work efficiently and dealing with problems like not having enough data. Our research seeks to enhance both adaptability and scalability in AI decision-making systems which is especially important for real-world uses and helps grow AI in web and edge areas.