Object-based classification of unmanned aerial vehicle (UAV) imagery for forest fires monitoring


38th Asian Conference on Remote Sensing - Space Applications: Touching Human Lives, ACRS 2017, New Delhi, India, 23 - 27 October 2017 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • City: New Delhi
  • Country: India
  • Recep Tayyip Erdoğan University Affiliated: No


© 2017 ACRS. All rights reserved.In recent years using of UAV has rapidly increased in assessment and monitoring of forest related studies. Due to higher spatial resolution of UAV images, make them possible to use for the extraction of forest and tree species with higher accuracy. However, the orthophoto map produced from Unmanned Aerial Vehicle (UAV) images must be evaluated and interpreted quickly. Different image classification methods are used for this purpose. Currently, object based image classification method which extracts data with high accuracy from high spatial resolution images has been used as popular classification technique. In this study, Camburnu Natural Park in Surmene District of Trabzon Province located in the Black Sea Region was selected as the study area. This area is unique in the world where yellow pine forests can land at sea level. An unpredicted forest fire realized on January, 8 2017 and around 20 hectares were destroyed. To determine destroyed area, current high resolution UAV images of study area were obtained and image pre-processing steps was employed. Object based classification is applied in two part as segmentation and classification. Firstly, multiresolution segmentation is processed for optimum parameters (scale, shape, color, compactness and smoothness) and segments are assigned proper classes by rule based. In this way images are classified by using object-based classification method. As a result, optimum parameter is determined for segments. The boundaries of forest and burned forest in the study area have been handled with high precision in vector format without salt and pepper effect.