부산대학교 도서관

UAV images are important data sources for search and rescue works both on sea and on land. It is a modern, suitable technological solution and in many situations there is no alternative method. Geometrical approach to pattern recognition tasks does not address spectral abnormal signatures in captured scenes. In this study the authors have investigated effectiveness of anomaly detection algorithms on different color spaces transformed from UAV images. The main goal of the research is to discover the most appropriate couples algorithm — color space in different terrain types. The proposal approach have been validated on test datasets and has shown promising results.