Infinite Latent Feature Selection Technique for Hyperspectral Image Classification

The classification process is one of the most crucial processes in hyperspectral imaging. One of the limitations in classification process using machine learning technique is its complexities, where hyperspectral image format has a thousand band that can be used as a feature for learning purpose. This paper presents a comparison between two feature selection technique based on probability approach that not only can tackle the problem, but also improve accuracy. Infinite Latent Feature Selection (ILFS) and Relief Techniques are implemented in a hyperspectral image to select the most important feature or band before applied in Support Vector Machine (SVM). The result showed ILFS technique can improve classification accuracy better than Relief (92.21% vs. 88.10%). However, Relief can extract less feature to reach its best accuracy with only 6 features compared with ILFS with 9.