One of the biggest challenges in an underwater target detection is that, unlike land-based scenes, the observed spectrum of an underwater target is highly dependent on the particular background that is in the scene. In particular, the observed spectrum is determined by not only the target reflectance signature but also by the optical properties of the water in which it is situated, as well as the depth of the target. It follows that signature-based detection algorithms must be able to accommodate the wide variation of observed spectra that a single target may exhibit in nature, and at any depth. In this article, we present a general framework for underwater detection in hyperspectral remote sensing imagery that uses physics-based modeling to calculate the target space—the set of all possible observed spectra for the target in a given scene—and then uses nonlinear mathematical models to exploit the structure intrinsic to the target space in order to reduce dimensionality and greatly simplify the detection problem. We include a series of simulated target images that demonstrates the effectiveness of this approach for multiple targets and depths.

中文翻译：

---

用于高光谱图像的水下目标检测框架

水下目标检测的最大挑战之一是，与陆基场景不同，水下目标的观测光谱高度依赖于场景中的特定背景。特别是，观察到的光谱不仅取决于目标反射率特征，还取决于其所在水体的光学特性以及目标的深度。因此，基于特征的检测算法必须能够适应单个目标在自然界和任何深度可能表现出的观察光谱的广泛变化。在本文中，我们提出了高光谱遥感图像中水下探测的通用框架，该框架使用基于物理的建模来计算目标空间——给定场景中目标的所有可能观测光谱的集合——然后使用非线性数学模型来利用结构目标空间固有的，以降低维数并大大简化检测问题。我们包括一系列模拟目标图像，展示了这种方法对多个目标和深度的有效性。