Locally adaptive spatial-spectral weighted hyperspectral image enhancement method for target detection (<italic>invited</italic>)

ObjectiveHyperspectral imaging delivers rich spectral information for precise perception but is hindered by high dimensionality and spectral redundancy. These limitations impose significant computational burdens and pose challenges to real-time target detection. Current enhancement methods are often optimized for generic image quality metrics, such as the Peak Signal-to-Noise Ratio (PSNR) or information entropy. However, since these objectives may not align with the requirements of detection tasks, such methods do not inherently improve target detectability. Consequently, a new enhancement paradigm is needed—one that directly maximizes the separability between targets and background, thereby offering more targeted support for reliable and efficient detection in practical applications.MethodsThe proposed method redefines the enhancement objective from overall image quality improvement to the maximization of target-background separability. It operates in two core stages: optimal band subset selection and adaptive image fusion. First, a multi-band joint Signal-to-Clutter Ratio (SCR), derived from the Mahalanobis distance, is introduced as a band selection criterion to directly quantify target saliency in multidimensional spectral space. An efficient bidirectional greedy search algorithm is designed to select a compact band subset with high information content and low redundancy. Subsequently, a locally adaptive spatial–spectral weighted Orthogonal Subspace Projection (OSP) fusion method is proposed. This method employs a dual weighting mechanism—based on spatial distance and spectral similarity—to estimate local background statistics for each pixel. This enables precise background suppression and directed enhancement of target signals, effectively addressing challenges in complex scenes and target edge regions.Results and DiscussionsExperiments were conducted on four diverse hyperspectral datasets: San Diego, PaviaC, MUUFL Gulfport, and WHU-Hi-River (pseudo-color composites and target ground truth are shown in Fig.3). The proposed method achieved consistent and significant performance improvements across all datasets. Quantitative evaluation using the SCR metric showed enhancements ranging from 16.06% to 779.94% compared to the best single-band images (Tab.1). For example, on the San Diego dataset, the SCR increased from 1.73 (best single band) to 15.24 after enhancement. Visually, the enhanced images (Fig.4(e)-4(h)) displayed sharper target contours and more effective background suppression compared to the best single-band images (Fig.4(a)-4(d)). A visual comparison between fusion results using the standard local OSP operator and the proposed spatial–spectral weighted OSP operator (Fig.2) confirms the latter’s superior ability to preserve target edges and handle heterogeneous backgrounds. These results validate the effectiveness of the joint SCR criterion for band selection and the robustness of the spatial-spectral weighting mechanism in complex environments.ConclusionsThis study introduces a novel hyperspectral image enhancement method specifically designed for target detection. Its main contributions are twofold. First, a multi-band joint SCR criterion and an efficient bidirectional greedy search algorithm are developed to systematically select an optimal band subset that maximizes target detectability. Second, a locally adaptive spatial-spectral weighted OSP fusion method is proposed, which integrates spatial and spectral weights to significantly improve background estimation accuracy and enhance target signals—particularly in complex scenes and near target boundaries. Experiments on multiple datasets demonstrate that the method substantially improves SCR and produces visually clearer target enhancements, confirming its effectiveness and generalizability. This work provides a more targeted and practical solution for real-time hyperspectral target detection. Future research may focus on more robust target spectral acquisition and adaptive parameter tuning mechanisms for the spatial-spectral weights.