Developing An Artificial Intelligence System For Tomato Stress Recognition Using Convolutional Neural Networks

Abstract

Convolutional Neural Networks (CNN), as artificial intelligence (AI) technology, is vital in identifying and recognizing features that affect cropping from images. Therefore, the current research uses CNN to diagnose tomato stress through thermal images. The paper presents four models of deep learning for image training: VGG16, Inception ResNet V2, Inception V3, and SVGGNet. The normal and Stress tomato leaf images taken from the field (thermal images) were trained on the four models. Transfer learning was used to train the VGG16, Inception ResNet V2, and Inception V3 models, while the SVGGNet model was trained from scratch. The dropout ratios of the models were studied at the ratios of 20, 40, 60, and 80% at an image size of 299 x 299 pixels. The most important results when training limited-size thermal images to diagnose thermal stress were the superiority of the SVGGNet model that was trained from the beginning over other models at all the dropout ratios that were studied. The best result was at a dropout of 40%, and its results came with an accuracy of 96.88% and a loss of 0.19. It has been shown that it is not preferable to train small-scale thermal data using transfer learning, but it is preferable to train the model from scratch.