An Improved Water Body Segmentation from Satellite Images using MSAA-Net

Satellite image water body identification is an important part of scientific progress. Because of factors such as aquatic vegetation, the unique lake or river hues, sediments along the bank, and so on, it can be difficult to determine where one body of water ends and another begins. To improve the process of extracting water bodies from satellite photos, both the variety of characteristics and the information they provide about their meaning need to be expanded. It may be utilized in a variety of contexts, such as forecasting the occurrence of natural catastrophes and identifying the presence of drought and flooding. As it relates to the study of remote sensing and image interpretation, automatically extracting water bodies from many satellite images with varying complexity targets is a crucial and challenging job. Mostly in the area of semantic segmentation of remote sensing images, convolutional neural networks, often known as CNNs, have emerged as a popular option in recent years. When it comes to executing water body segmentation, however, generic CNN models face a number of challenges. In this paper, in order to locate the bodies of water, we utilized the MSAA-Net deep CNN that was hosted on TensorFlow. The dataset downloaded from Kaggle was used to train the model that was suggested here. It may be broken down into two categories: water bodies and masks. There are a total of 2841 photos captured by the Sentine1-2 satellite, along with 2841 associated masks. The findings indicate that our technique provides the highest possible level of segmentation performance, outperforming both the traditional and more contemporary methods. In addition, the technology that we have presented is resilient in difficult extraction circumstances, including aquatic bodies.