Application of Hyperspectral Image with AI Models for Monitoring of Coastal Ocean and Land Surface

Remote sensing is a powerful technique of classifying and quantifying objects for the coastal waters and land surfaces. However, the elaborate classification of the ultra-high-resolution image is still challenging in the coastal regions due to high possibility of class mixing in coastal waters with complex structures. The hyperspectral images have advantages in precise work due to high spectral resolution in a wide bandwidth. This study introduces the case study result using a novel method to classify green algae on artificial structure based on hyperspectral data and the deep learning models. The spectral characteristics of the attached green algae was observed using a ground-based hyperspectral camera. The attached green algae show a spectrum of different features, depending on the distribution density. Subsequently, 2-D waveform images were produced representing spatial and spectral characteristics for individual pixels. We performed the dense convolutional network (DenseNet) and basic convolutional neural network (CNN) models to classify green algae into three levels which give 97.94 % and 97.20 % classification accuracies, respectively. In particular, both models showed superior classification performance compared to the SVM model at the concrete and algae interface.