Exploring Advanced Techniques for Agricultural Area Mapping: Comparative Analysis of U-Net and Mask-Based Approaches in Satellite Imagery

Abstract

Deep learning methodologies have rapidly gained prominence in the realm of statistical analysis for remote sensing data. These techniques are instrumental in processing spectral information, conducting identification statistics, and performing segmentation and classification of objects in satellite imagery. Image segmentation enhances the accuracy of object statistics by effectively distinguishing objects from their backgrounds. This paper delves into the application of Mask Based Approaches and U-Net in satellite imagery segmentation. An experimental study  is conducted to evaluate the suitability of these models in this domain. The experimental results, measured in terms of mean average precision (mAP), indicate a remarkable performance, with a precision of 93.21% for Mask-Bcoased Approaches and 90.69% for U-Net, based on a dataset comprising satellite images of Pakistan. These findings underscore the potential of these techniques in advancing the accuracy and efficiency of object segmentation in satellite imagery analysis.