Transfer Learning-Based Approach for Citrus Disease Detection

Abstract

Fruit diseases have a devastating impact on the food economy, wreaking havoc on the global food business. Therefore, in agriculture and food security, disease detection and diagnosis systems must be efficient and dependable. Technology supported by computers can be used to detect illnesses. Convolutional neural network (CNN) approaches have recently demonstrated impressive performance in image categorization applications. CNN can operate relatively efficiently with high-volume datasets and extracts more circumstantial features. This paper offers a strategy that employs deep convolutional neural networks with transfer learning to identify and categorize citrus crop diseases. Useful features were identified from fine-tuned, pre-trained deep learning models that were retrained using an image dataset of citrus fruits with infections. On a large dataset (ImageNet), the InceptionResNetV2, DenseNet169, MobileNet, NasNetLarge, NASNetMobile, and VGG19 models have been pre-trained to increase prediction accuracy. The VGG19 model has a classification accuracy of 97.54 percent, which is superior to current methods.