An Enhanced Spatio-spectral Template for Automatic Small Recreational Vessel Detection

This paper investigates the performance of a spatio-spectral template on Ikonos imagery to automatically detect small recreational boats. The spatio-spectral template is employed which is then enhanced through use of a weighted Euclidean distance metric adapted from the Mahalanobis distance metric. The purpose is to assist the Canadian Coast Guard in gathering data on recreational boating for the modeling of search and rescue incidence risk. To test the detection accuracy of the enhanced spatiospectral template, a dataset was created by gathering position and attribute data for 53 recreation vessel targets purposely moored for this research within Cadboro Bay, British Columbia, Canada. The Cadboro Bay study site containing the targets was imaged using Ikonos. The overall detection accuracy was 77 percent. The targets were broken down into two categories: Category A-less than 6 m in length, and Category B-greater than 6 m long. The detection rate for the Category B targets was 100 percent, while the detection rate for the Category A targets was 61 percent. It is important to note that some Category A targets were purposely selected for their small size to specifically test the detection limits of the enhanced spatiospectral template. The smallest target detected was 2.2 meters long and 1.1 meters wide. The analysis also revealed that the ability to detect targets between 2.2 meters and 6 meters long was diminished if the target was a dark color.