Methodology for Classifying Objects in High-Resolution Optical Images Using Deep Learning Techniques

The classification of objects that are present in the images or in the videos is being developed progressively to obtain good results because of the use of convolutional networks. In this work, we have used the convolutional networks for the detection of objects that are present in high-resolution satellite images. Tests were carried out on ships that are on the high seas and in the ports. This classification is useful for monitoring the coasts, as well as for analyzing the dynamics of the ships which can be applied in the search of ships. To cover this task of classifying ships in the spectral images, the use of high-resolution satellite images of coastal areas and with a large number of ships is used in order to build a set of images, containing images of the ships. In order to be used for training setting and testing of the convolutional network, a very particular configuration of the convolutional network caused by the particularity of high-resolution satellite images is presented. The methodology developed indicating the procedures performed is also presented in which a set of images containing 300 was built images of ships that are in the sea or are anchored in the ports. The results obtained in the classification using the convolutional networks are acceptable to be able to be used in different applications.