A Neural Network-Assisted GLRT Detection Algorithm for Range-Spread Target

The range-spread target detection problem typically faces uncertainty in the number and locations of target scattering centers (TSCs), which severely limits the performance of traditional algorithms. Conventional methods estimate the TSCs based on the energy information of echo signals and heavily rely on certain assumptions and specific prior knowledge. The performance of these algorithms may degrade significantly in the presence of assumption mismatches or a lack of required prior information. In practical applications, echo signals from one target shall exhibit some aspects of correlation, while clutter echo signals from different range cells are independent. This distinct characteristic between target and clutter may offer extra information in range-spread target detection. Therefore, a neural network based on the intrinsic features of echo signals is proposed for TSCs estimation by outputting the probability of TSC presence in each range cell. In addition, conventional methods directly sum the test statistics of estimated TSC-present range cells to form the final test statistic. This approach may lead to performance degradation, as weak target signals diminish the contribution of strong target signals to the integrated test statistic. To address this limitation, we propose to assign weights to the test statistics of each range cell based on the predicted probability. This approach enables more effective integration of the TSCs energy of all range cells under test. Simulation experiments demonstrate that the proposed algorithm achieves substantial performance improvements over the traditional methods.