Color is probably the most informative cue for object recognition and classification in natural scenes. Difference in shades can indicate to the biologist the potential for diversity of species or stress on the habitats. However, severe color distortions may occur in underwater imagery due to wavelength-dependent attenuation of light. Affordable tri-chromatic sensors are used to record the ambient light condition and color correct the imagery, but results show that this approach works reliably only under highly controllable conditions. This paper proposes an approach that combines hyperspectral data collected for the object of interest, hardware properties of the imaging sensor, and exterior conditions (optical properties of water and illumination) with tri-chromatic underwater imagery. Due to ambiguity of color reconstruction underwater, demonstrated in the paper, a probabilistic approach is used for classification that allows the identification of the object of interest from other objects.