Evaluation of a U-Net Based Method for Segmenting Unauthorized Airstrips in the Amazon Rainforest using SAR Imagery

In recent times, the Amazon Rainforest in Brazil has faced significant environmental damage as a result of illegal human activities often facilitated by unlawful landing strips. These activities include mining and drug trafficking. This research presents a method for detecting these airstrips in 16- bit per pixel synthetic aperture radar (SAR) images using a U-Net model. The rainforest is extensive and known for its dense vegetation. The initial phase involved creating a database of SAR images from the Sentinel-1 satellite in their original TIF format. The database included 646 images for training, 277 for validation, and 117 for generalization testing. The findings suggest that the U-Net network can effectively detect features corresponding to the airstrips and accurately segment the targets in synthetic aperture radar images with 16 bits.