 This paper investigates the use of deep convolutional neural networks, CNNs, for the classification of complex wetland classes using multi-spectral rapid eye optical imagery. It compares the performance of seven popular CNN architectures, DenseNet 121, Inception V3, VGG16, VGG19, Seption, ResNet 50, and Inception ResNet V2 against two traditional machine learning algorithms, Random Forest and Support Vector machines. The results show that the full training of Inception ResNet V2 provides the best accuracy of 96.17%. This is significantly higher than the other CNN models tested, which range between 74.89% and 93.57%. Additionally, it is also better than the two traditional machine learning algorithms which achieve accuracies of 74.89% and 76.08%. These findings suggest that Inception ResNet V2 is the most effective CNN model for classifying complex remote sensing scenes such as wetlands. This article was authored by Masood Mahdi Amperi, Baram Salehi, Muhammad Rezee, and others.