Hybrid Learning-Based Region Growing for Remote Sensing Image Segmentation

Remote-sensing is now a days, an indispensable instrument for the applications like environmental monitoring and land use analysis. This paper presents a novel approach combining semi-supervised region-growing with self-supervised learning in order to enhance image-partitioning in remote-sensing applications. Our method leverages, strengths of both techniques: region-growing effectively parts imageries based on local pixel similarities, while self-supervised adaptive training derives meaningful characteristics depictions from vast amounts of unlabeled data. We integrate said methodologies to improve the efficiency of segmentation tasks with accuracy, even with limited labeled data. The proposed approach utilizes a self-supervised autoencoder to learn robust features, which guide the region-growing algorithm, enabling it to adaptively refine segment boundaries. Extensive experiments on satellite imagery datasets demonstrate significant improvements in segmentation performance, evaluated using metrics such as Intersection over Union (IoU) and Dice coefficient. Qualitative assessments further confirm the enhanced precision of our method. This work illustrates the potential of hybrid techniques in remote sensing and provides a foundation for future advancements in semi-supervised learning for image processing.