Enhanced affinity propagation clustering with a modified extreme learning machine for segmentation and classification of hyperspectral imaging

Abstract

Hyperspectral Imaging (HSI) plays a crucial role in detecting, identifying, and classifying a wide range of natural resources, including minerals, geological phenomena like volcanic eruptions, and vegetation. Segmentation and classification of HSI play vital roles in extracting meaningful information and identifying different land cover or land use categories within the scene. One of the primary limitations associated with HSI is the scarcity of labeled samples. Obtaining annotated samples is a laborious and time-consuming process, posing a significant challenge in the field. This work presents an Enhanced Affinity Propagation Clustering (EAPC) and Modified Extreme Learning Machine (MELM) for segmentation and classification of HSI. Initially, the HSI images are pre-processed by the non-linear diffusion partial differential equation. Then, the segmentation process is performed by the EAPC and it is the combination of Affinity Propagation Clustering (APC) with Light Spectrum Algorithm (LSA). Finally, the classification is performed by the MELM and the experimentation is demonstrated on the Salinas dataset and achieved better accuracy and sensitivity of 97.3 % and 98.2 % respectively.

Keywords

• Hyperspectral imaging
• Enhanced affinity propagation clustering
• Classification
• Modified extreme learning machine