Multi-sensor image fusion and target classification for improved maritime domain awareness

In this paper, we propose a scheme for classification of maritime targets through fusion of images collected from dissimilar sensors with an objective to improve maritime domain awareness. Low- and medium-level fusion methods are applied to three types of image data-visual, thermal, multi-spectral-using features obtained from the speeded-up robust features algorithm. The goal was to implement the classification scheme using machine learning techniques. Results indicate that multi-spectral images from low-level fusion yielded the best classification performance. Artificial neural networks are used to derive the classification results and demonstrate the ability to obtain results in a timely manner that could accommodate near real-time classification.