Soil Study of Coastal Hyperspectral Data using K-means and LDA (Latent Dirichlet Allocation)

Recent advances in Hyperspectral imaging have induced the growth of various paradigms for advanced systems in remote sensing and Geospatial Data Analysis. Various hyperspectral sensors have been laden on artificial satellites or UAV flights, to capture granular data regarding the topologies of the land. Unsupervised learning is generally used to classify and identify portions of an image. In this paper, the effectiveness of the K-Means Clustering Algorithm and Latent Dirichlet Allocation (LDA) in clustering the soil type present on a beach is checked. The dataset used is from high signal-to-noise ratio imaging spectroscopy (AVIRIS-NG). The data was visualized using the ENVI toolkit present in Python's Spectral library. Clustering was performed, and an analysis was conducted about the similarity in the same. The soil types are classified, and the two algorithms are contrasted and effectiveness is generalized.