A systematic review of hyperspectral underwater target detection under complex water conditions was presented from three perspectives: imaging mechanism, characteristic modeling, and algorithm design. Starting from the underwater hyperspectral imaging mechanism, existing methods were categorized into five groups: spectral prediction, spectral restoration, band selection, pixel classification, and feature construction. Their differences and connections were compared in terms of mechanism-consistent modeling, distortion correction, representational robustness, and interpretability. Analysis show that current methods exhibited distinct characteristics in prior dependency, information utilization, and cross-scene adaptability, and the technical approaches are evolving from mechanism-oriented analysis toward mechanism-data synergy, as well as the integration of generative modeling and feature construction. On this basis, the major challenges in environmental adaptability, reliability modeling, and generalization are further summarized, and future directions are discussed, including differentiable physical modeling, uncertainty characterization, and cross-scene generalization mechanisms.