Water Detection in Satellite Images Based on Fractal Dimension

Identification and monitoring of existing surface water bodies on the Earth are important in many scientific disciplines and for different industrial uses. This can be performed with the help of high-resolution satellite images that are processed afterwards using data-driven techniques to obtain the desired information. The objective of this study is to establish and validate a method to distinguish efficiently between water and land zones, i.e., an efficient method for surface water detection. In the context of this work, the method used for processing the high-resolution satellite images to detect surface water is based on image segmentation, using the Quadtree algorithm, and fractal dimension. The method was validated using high-resolution satellite images freely available at the OpenAerialMap website. The results show that, when the fractal dimensions of the tiles in which the image is divided after completing the segmentation phase are calculated, there is a clear threshold where water and land can be distinguished. The proposed scheme is particularly simple and computationally efficient compared with heavy artificial-intelligence-based methods, avoiding having any special requirements regarding the source images. Moreover, the average accuracy obtained in the case study developed for surface water detection was 96.03%, which suggests that the adopted method based on fractal dimension is able to detect surface water with a high level of accuracy.

Automated summary

Identification and monitoring of existing surface water bodies on Earth are important in various scientific disciplines and industries. This study establishes and validates an efficient method for surface water detection using high-resolution satellite images and data-driven techniques. The method, based on image segmentation, Quadtree algorithm, and fractal dimension, shows clear distinction between water and land, and it is simpler and computationally efficient compared to heavy artificial-intelligence-based methods. The average accuracy obtained in the case study was 96.03%, indicating high accuracy of the adopted method based on fractal dimension.