Ship detection by different data selection templates and multilayer perceptrons from incoherent maritime radar data
This study presents a novel way for detecting ships in sea clutter. For this purpose, the information contained in the Radar images obtained by an incoherent X-band maritime Radar is used. The ship detection is solved by feedforward artificial neural networks, such as the multilayer perceptrons (MLPs). In a first approach, the MLP processes the information extracted from the Radar images using the commonly used horizontal and vertical selection templates. But, if a suitable combination of these selection templates is done, better detection performances are achieved. So, two improved selection templates are proposed, which are based on cross and plus shapes. All these templates are also applied in a commonly used detector taken as reference, the CA-CFAR detector. Its performance is compared with the one achieved by the proposed detector. This comparison shows how the MLP-based detector outperforms the CA-CFAR detector in all the cases under study. The results are presented in terms of objective (probabilities of false alarm and detection) and subjective estimations of their performances. The improved MLP-based detector also presents low computational cost and high robustness in its performance against changes in the sea conditions and ship properties.
