A Multiple Mechanism for Identifying Green Land Cover Using Auto Encoder in Satellite Image Processing

Abstract

The automatic identification of green land cover zones in satellite images may be crucial for preserving ecological balance and guaranteeing environmental safety after deforestation. With this research, we hope to answer the demand for an automated tool that may serve as an assistive ecosystem service provider by developing a novel two-stage framework for identifying green land cover in satellite images. The framework is specifically designed to tackle the challenges related to the expanding application of automated computer vision-based frameworks for green land cover detection based on satellite imagery. Initially, distinct kinds of chromatically homogenous land cover areas in the satellite picture are segmented using a clustering approach. To isolate only green land cover clusters from other clusters that signify water bodies, arid regions, built-up areas, etc., an auto encoder model is constructed in the second stage using a range of green land cover samples. The proposed framework shows an impressive over 97.5% detection accuracy even with a small experimental dataset, which is comparable to the results of widely used Mask R-CNN based segmentation techniques.