Automated Channel-Network extraction from satellite images using deep convolutional neuralNetworks : a case study of Mun river in Thailand

Abstract

Main river and oxbow lake are significant factors in Thai agriculture for producing agricultural production. The National Statistical Office Thailand stated that in 2018, agriculturists, which has the percentage of 35 of the total labor forces or 12.37 million from 38.26 million face, lack income due to a low productivity from the drought or flood caused by the inefficient water resource management. In the present day, the study of water resource management by using satellite images has three methods. First of all, using human digitization by ArcGIS or QGIS. The disadvantages of this method are that it has high human error and is time consuming. Secondly, using water index (NDWI) to extract water bodies from satellite images. Lastly, using machine learning to cluster the objects into each class. The weakness of the second and third methods is that there are water bodies beyond the study scope and solid bodies disrupt the specific area identification. Therefore, this study applies deep convolutional neural networks to extract Main river and Oxbow lake specifically together with calculating water surface highly accurately and automatically to improve the water resource management in local areas rapidly and efficiently.