Comparative Analysis of CNN Models' Accuracy Using Synthetic and Original Datasets for Cotton Leaf Classification

In agriculture, finding diseases in plants as soon as possible is very important for getting the most crops and making sure there is enough food for everyone. A potential way to quickly and accurately diagnose diseases is to use Artificial Intelligence (AI) tools, especially image classification models. But the quality and variety of the training sample have a big impact on how well these models work. In many cases, it's hard to get a fair collection with enough samples. In order to solve this problem, we suggest a new way to classify cotton leaf diseases that uses both fake and real datasets. First, we use the Deep Convolutional Generative Adversarial Network (DCGAN) method to make a fake dataset. This makes fake pictures that look like the ones in the real dataset. After that, we use both the real and mixed datasets to train different Convolutional Neural Network (CNN) models, such as DenseNet121, InceptionV3, MobileNetV2, ResNet50, VGG16, and VGG19. To see how well each model does on both sets of data, evaluation measures like accuracy, Fl score, and classification results are used. Our results show that the accuracy rates of the combined dataset with simulated pictures are higher than those of the original dataset alone.