Deep Learning for Sustainable Farming: Multi-Class Disease Detection in Apple and Tomato Leaves

Improving agricultural production and overall food quality requires accurate identification of diseases affecting plants. Traditional disease detection techniques are time taking and require specific knowledge, making them unsuitable for widespread agricultural observation. An automated approach for disease classification is presented in this paper. A research study is conducted using a dataset of apple and tomato plant leaf images to compare seven different deep learning architectures: EfficientNet, ResNet50, DenseNet121, MobileNet, AlexNet, Xception, and GoogLeNet. In terms of performance and relevance in limited resource situations, GoogLeNet performed better than other models with a validation accuracy of 95.84%. While ResNet50 and EfficientNet-B0 have shown comparable results, we have chosen GoogLeNet as the preferred model, considering its computational efficiency and memory usage. To increase the model's generalization ability, several data augmentation methods such as image scaling and normalization are applied after the pre-processing of data. The future study initiatives will focus on increasing the dataset while also optimizing the model for better accuracy across a greater variety of agricultural diseases.