Hyper-spectral remote sensing water depth retrieval based on spectral difference factors

Many studies have indicated that spectrum is mainly decided by substratum and water depth in shallow water, so spectrum above one kind of substrate is only decided by water. According to this idea we studied the technology of substrate classification, as well as analyzed the impacts of various water-depth extraction factors on the inversion accuracy. The following results have been obtained. (1) SVM has the highest classification accuracy, whose Kappa coefficient was 0.86 and overall accuracy was 92.34%, which is higher than that of neural network and maximum likelihood. (2) Correlation coefficient between factors based on spectral shape and water depth were over 70%, which is higher than that based on spectral amplitude. (3) SA and SGA are all have an exponential correlation with water depth and their inversion accuracy was almost the same. The mean relative error and mean absolute error for two factors were 9.9%, 0.61m and 7.3%, 0.74m, respectively. But they have different performance in various substrate area and depth.