Effectiveness Of Pre-Trained Deep Learning Models In Smart Agriculture For Leaf Disease Detection

Abstract

With major crop losses and jeopardising food security, tomato diseases seriously confront world agriculture. Reducing the economic and environmental consequences of these illnesses depends on fast and accurate diagnosis of them. This work investigates the use of deep learning method known as transfer learning to enhance tomato leaf disease diagnosis. Transfer learning, a technique in deep learning, allows models to leverage pre-trained knowledge on similar tasks, improving performance and reducing the need for large datasets. This method has gained popularity in agricultural disease detection due to its efficiency and adaptability across different domains We investigate many pre-trained models including AlexNet, DenseNet, ResNet18, VGG16, VGG19, and InceptionV3 in their ability to identify certain tomato illness. Key performance measures including accuracy, precision, recall, computational efficiency, and resource consumption guide evaluation of the models. AlexNet has the greatest overall performance, according to the results, which mix accuracy with computational economy to make it particularly appropriate for settings with low resources.  Conversely, deeper models like ResNet18 and VGG16  need large processing effort but have better accuracy. These results show that transfer learning may be a powerful tool for better agricultural disease diagnosis, which can lead to better crop management methods, more accurate disease detection in tomato leaves, and ultimately, more food security and more sustainable farming.