Detection of Water Bodies from Satellite Images: A Deep Learning Based Technique

One of the interesting areas of use of remote sensing is in the observation of surface water and detection of water bodies. for studying ecological and hydrological processes. However, detection of small water bodies from satellite images is challenging. In this work, a deep-learning based approach is proposed to detect water bodies and mask the areas in the images. The proposed approach aims to detect water bodies of different shapes and sizes including small water bodies and those with vegetation. A large number of satellite images is taken from Indian Space Research Organization's (ISRO) Cartosat – 3 satellite and is annotated to prepare the labelled samples. Thereafter, the model is prepared and trained till the loss saturates. The model when tested on further set of images, detects clear water bodies and also those with vegetation. In all cases, the trained model is able to effectively mask the boundaries of water bodies in test images with a mean precision of 89% and accuracy over 95%. Along with the masked image, the total number of water bodies and their area are also calculated in the output.