FruVeg-Net: A Novel Method for Early Disease Diagnosis in Multi-fruits and Vegetables

Fruit and vegetable pathogens have a catastrophic impact on food safety and can dramatically reduce the value and potency of agricultural products, which may destroy the entire farmland in extreme situations. Therefore, an autonomous pathogen detection and diagnosis system is necessitated in the fruitage domain. Nowadays, deep learning methods have been favored due to their exceptional results. However, the efficacy of extracting the minute lesion feature of fruits has been compromised, resulting in poor precision. To address the same, a deep convolutional neural network-based transfer learning model FruVeg-Net is presented for identifying and classifying multiple fruit diseases. Different state-of-the-art architectures like InceptionV3, ResNet50, and VGG19 models have been pre-trained on the huge database (ImageNet) to improve recognition accuracy, which is further enhanced using augmentation techniques. A variety of performance metrics like accuracy, loss, and ETA for training, validation, and testing datasets are measured to compare accuracy on four datasets like tomato, guava, citrus, and mixed fruits. Our experimental findings indicate that the InceptionV3 model overshadows the other models with the highest recognition accuracy, 99.34%, for the mixed fruit dataset.