Fast marching and acoustic descriptors based method for fish proportion interpolation

We propose a new method for the estimation of fish abundance from both acoustic data and some trawl hauls catches. In this work, we operate at a global level and we aim at estimating fish abundance from these images and not to identify the species of each school. We associate each trawl catch to the nearest acoustic image and we describe each image by a set of global statistical distributions estimated on it and related to each school parameters. Then, we use the fast marching algorithm, a front propagation based scheme to define a region of interest around each trawl associated image. The fast marching algorithm propagates each front initialized on the image associated to the trawl samples with a velocity proportional to the distance between the trawl image acoustic features and those of the image for which we want to estimate the fish abundance. Finally, the fish abundance of each image is estimated as a weighted sum of the abundances associated to each trawl. The weights are estimated from the propagation time given by the fast marching algorithm. The method is experimented on real and synthetic data.