Arial Image Classification using Deep Neural Networks with Discrete Cosine Transform, TSBTC and Augmentation Techniques

The surfacing of continuous monitoring needs, dramatic increases in computer processing power, enhanced modes of connectivity, processing of images, and capacity to capitalize on information in images are some of the reasons for the emergence of aerial vehicles. The major challenge in this field has been segregation of the aerial images captured by these vehicles. Several attempts have been made for successfully achieving good results for classification of aerial images using K-Nearest Neighbors, decisions trees, image segmentation, Support Vector Machines and k-means clustering. This paper attempts the same using Deep Convolutional Neural Network (DCNN) with an integration of Thepade’s Sorted Block Truncation Coding n-ary (TSBTC nary) and Cosine Transform of aerial images from the UC Merced Dataset. A comparison of the performances using these methods is done and it is concluded that TSBTC n-ary on Discrete Cosine Transformed (DCT) images with augmentation gave the best results. The TSBTC 4-ary augmented model gave outstanding results, than most modern methods, with an accuracy of 99.29%.