The abundant spectral information provided by hyperspectral images (HSIs) greatly benefits target detection (TD), but it home introduces a large amount of data redundancy, which greatly increases the complexity of data processing.Band selection (BS) is an important preprocessing step for HSI applications, and it plays an important role in reducing data volume and improving processing efficiency.Compared with BS methods that do not have specific applications, BS methods with specific applications often have clear objectives and better band selection results.Based on this idea, a series of BS methods called target feature constrained-based (TFC) used for the field of TD are developed and proposed in this paper.These methods can select the band subset that simultaneously enhance the desired targets and suppress the undesired targets.
Its idea is derived from the target-constrained interference-minimized filter (TCIMF).By analyzing the detection principle of TCIMF, the BS problem can be transformed into a constrained optimization problem.According to the different method for solving constrained optimization problem, the TFC methods are divided into TFC-MP (TFC-Matching Pursuit), TFC-OMP (TFC-Orthogonal Matching Pursuit) and TFC-StOMP (TFC-Stagewise Orthogonal Matching Pursuit) methods.The experimental results show that, educational toys compared with several traditional BS methods, the TFC methods achieve better accuracy, efficiency and performance.