Skip to Main Content
This study presents the problem of detecting known targets (Swerling 0 model) in simulated ground clutter (land cultivated). As a modelled low-resolution coherent radar system is used, the clutter is modelled by a complex-valued time-correlated Weibull distribution. The research exposed in this study looks for two objectives. First, finding a detection scheme that permits the final user (radar operator) determine if a target is present or not, and its position, size and shape. And second, designing a detection scheme as robust as possible against clutter and target condition changes and approximating the specification and measurement of the radar performance by using clutter models. However, these tasks are really complicated, because high-level clutter echoes are received. A detection scheme based on neural networks is proposed, where feedforward multilayer perceptrons are used. The performance obtained with this detection scheme is discussed from subjective (visual analysis of the achieved detection scans) and objective (receiver operating characteristics) points of view. Finally, this scheme is compared with a coherent detector commonly used when Weibull-distributed clutter is present, the target sequence known a priori detector. This comparison empirically demonstrates the performance improvement achieved by the proposed detector comparing with the commonly used.