Random Forest based Coral Reef Health Classification Using Satellite Imagery and Geographical Information System

Understanding the coral reefs’ health in varied climate changes is challenging with the traditional methods. To address this, the proposed research developed a Machine Learning (ML) framework for effective classification using Geographical Information System (GIS). The data in this research, is collected from Great Barrier Reef (GBR) in Australia which includes the satellite Landsat 8 images for monitoring coral reefs condition. The data is pre-processed using normalization and scatter plots to analyse various coral reef classes using the training samples. The images features are extracted using Gray-Level Co-occurrence Matrix (GLCM) and analysed the spectral indices to capture specific type of information about the coral reefs. The coral reef images are then segmented using Pixel Based (PB) and Object Based (OB) segmentation to achieve effective Coral Reef Health Classification (CRHC). To analyse the large amount of data with the non-linear relationships between the variables, the Random Forest (RF) classifier is used. The experimental results showed that the proposed RF-CRHC achieved better accuracy with 99.59% and precision with 97.79% when compared to the existing methods Multi-Local Perception Network (ML-Net) and Clustering based RF.