Multiclass Vessel Detection From High Resolution Optical Satellite Images Based On Deep Neural Networks

One of the core components of remote sensing based maritime surveillance applications is vessel detection. It helps to prevent and investigate different unlawful actions as well as environmental hazards present at sea. Growing constellation of very high resolution (VHR) optical satellite sensors are able to frequently cover large areas with spatial resolution of up to 0.3m per pixel, which is sufficient to detect and distinguish different types of vessels. This paper presents a novel method for automatic multiclass vessel detection with use of deep convolutional neural networks (DCNN) and principle component analysis (PCA). The described approach provides reasonable performance and therefore is potentially suitable for near real time (NRT) applications