Improved hyperspectral point target detection via segmented matched filter and adaptive cosine estimator fusion

Point target detection in hyperspectral images (HSIs) is a critical yet challenging task in remote sensing, particularly due to complex background interference, spectral variability, and low signal-to-noise ratios. In this study, we propose novel detection strategies that leverage segmented versions of the Matched Filter (MF), a correlation-based spectral detection algorithm, and Adaptive Cosine Estimator (ACE), a similarity-based spectral angle detector, to improve target discrimination, to improve target discrimination. We introduce three fusion approaches: (1) Multiplication, which combines the outputs of segmented MF and ACE multiplicatively to amplify target signatures and suppress background noise; (2) Parent-Child, where one detector serves as a guide to refine and constrain the other's results based on spatial-spectral context; and (3) Mean + K×Std, an adaptive thresholding technique based on the local statistical distribution of detection scores within each segment. Our experimental evaluations, conducted on representative HSI datasets, demonstrate that all three strategies consistently outperform traditional MF and ACE techniques, achieving higher detection accuracy and improved robustness to background clutter. The segmentation approach enables better local adaptation, enhancing sensitivity to weak or partially obscured targets. These findings suggest that segmentation-based fusion strategies offer a promising new direction for hyperspectral point target detection and can be extended to other spectral analysis tasks.