MARINE CARBONATE SAND LOCATION AND SUBSTRATE MORPHOLOGY ANALYSIS USING PCA AND NEURAL NETWORKS ON RGB IMAGES

Though sand is an important aspect of near shore geology there is only limited understanding of its behavior and characteristics as a valuable natural resource, an integral component of the reef system, and a highly dynamic substrate. To facilitate understanding temporal variability and spatial distribution of sandy substrate, simple and accurate methods for sand identification in aerial photographs and digital images are essential. Principal Component Analysis of blue and green bands provides two new channels of information. Using the second eigenchannel provides increased distinction of bottom types, though not enough to accurately classify the image. Blue, green, and the second eigenchannel are used as inputs into the Artificial Neural Network (ANN), which assigns each pixel to an information class. ANN’s are effective and efficient tools for rapid segregation of basic substrate information classes, which in this case we reduce to two, thereby focusing our query on sandy substrate. The result of ANN classification is a two information class data set, with the primary information class identifying sandy substrate. Additional querying of the classified image using Fourier spatial analysis helps to define spatial characteristics of sand within the image. Finally, combining the sand information class with DEM data and field site surveys produces a detailed description of both sand and basin substrate characteristics. These methods combine to quantify sand occurrence as a component of the Waikiki carbonate reef system.