Hyperspectral imaging provides rich spectral and spatial information and has become an important means for underwater target detection. Compared with conventional visible or single-modality imaging, it offers stronger material discrimination capability and greater potential for identifying submerged targets. However, underwater hyperspectral target detection remains highly challenging because target observations are strongly affected by absorption, scattering, illumination fluctuation, and complex background interference. These factors lead to spectral distortion, weak target responses, reduced target-background separability, and unstable detection performance. In recent years, with the development of unmanned aerial vehicle platforms and hyperspectral sensors, unmanned aerial vehicle-borne hyperspectral underwater target detection has emerged as a promising research direction. This task not only involves underwater optical degradation, but also faces additional constraints such as platform motion, radiometric fluctuation, and limited onboard computing resources. Therefore, a systematic review of the imaging mechanism, characteristic modeling, and algorithm design in this field is of clear significance for understanding current progress and guiding future research.

This paper reviewed the research progress of hyperspectral underwater target detection from three perspectives: imaging mechanism, characteristic modeling, and algorithm design. Starting from the underwater hyperspectral imaging process, the paper first discusses the physical basis of spectral formation and explains how water absorption, scattering, optical path variation, and background coupling affect the observed spectral signatures of submerged targets. This analysis clarified why underwater hyperspectral target detection was fundamentally more difficult than land-based hyperspectral target detection and why mechanism-aware analysis remains important. Based on this foundation, existing methods were categorized into five groups: spectral prediction, spectral restoration, band selection, pixel classification, and feature construction. Spectral prediction methods mainly infer target-related spectral responses under underwater conditions by using prior information or predictive mappings. Spectral restoration methods attempted to compensate for water-induced degradation and recover more effective target spectra. Band selection methods identify informative bands to reduce redundancy and suppress disturbance, but they may lose useful discriminative cues when target information was distributed across multiple bands. Pixel classification methods directly establish decision boundaries between targets and background at the pixel level and can achieve good performance under relatively consistent data distributions. Feature construction methods focus on building more discriminative representations before detection, so that targets and background can be better separated in the learned feature space. The review further compared these five categories in terms of mechanism consistency, distortion correction capability, representation robustness, interpretability, prior dependency, and cross-scene adaptability. Existing studies showed that these methods were complementary rather than mutually exclusive. Mechanism-oriented methods usually have stronger interpretability, but their effectiveness may depend on simplified assumptions or incomplete environmental information. Data-driven methods have better flexibility in dealing with nonlinear and complex observations but may face limitations in reliability and generalization under scarce data or large scene variation. Recent progress indicates that the field is gradually evolving from pure mechanism-oriented analysis toward collaborative paradigms integrating physical priors, generative modeling, and feature construction. This trend reflects the need to balance interpretability, robustness, and adaptability in complex underwater environments. The paper also discussed the current status of datasets and evaluation practices. Existing underwater hyperspectral datasets differ considerably in water conditions, target types, acquisition platforms, spectral ranges, and annotation quality, which makes direct comparison among different methods difficult. Nevertheless, current studies have already revealed clear methodological trends and provided valuable understanding of the relationship between underwater imaging characteristics and algorithm design.

Hyperspectral underwater target detection is a rapidly developing but still challenging research area. Its main difficulties come from physical degradation, complex environments, limited prior reliability, and insufficient cross-scene generalization. Through a unified review framework, this paper shows that the field has evolved from early physical mechanism analysis to characteristic correction and restoration, and then further toward higher-level feature construction and mechanism-data collaborative modeling. This evolution suggests that future progress will rely more on the integration of physical understanding and robust learning strategies than on isolated improvements in single modules. Several future directions deserve particular attention. Differentiable physical modeling is expected to connect underwater imaging constraints with learnable frameworks more effectively. Uncertainty characterization is important for improving prediction reliability and supporting trustworthy decision-making in dynamic underwater environments. Cross-scene generalization mechanisms are also critical because practical applications require stable performance under varying water quality, illumination, target state, and acquisition conditions. In addition, future studies should strengthen benchmark construction, standardized evaluation, lightweight deployment, and interpretable model design. Overall, the continued development of hyperspectral sensing, underwater physical modeling, and intelligent detection algorithms will further improve underwater target detection capability and provide stronger support for high-precision and high-reliability underwater remote sensing applications.