This article presents the optimization and hybrid parallelization of a spatial-spectral feature extraction (FE) method from hyperspectral images (HSIs) using local covariance matrix (CM) representation, exploiting hybrid parallelism through multicore and manycore processors. The aim is to evaluate the performance of parallel versions of this innovative algorithm that characterizes spatial-spectral information prior to classification when conducting FE. The HSI is first projected into a subspace, using the maximum noise fraction method. Then, for each test pixel, its most similar neighbors are clustered using the cosine distance measurement. The result is used to calculate a local CM with each nondiagonal entry characterizing the correlation between different spectral bands. Such matrices represent the spatial-spectral features and are fed to a support vector machine for classification. To optimize the successive parallelization process, a new version of the original MATLAB code has been first developed using C language. This serial version serves as baseline for hybrid parallelization in OpenMP and CUDA. Performance analysis has been conducted using publicly available HSI datasets, confirming that our parallel implementation ensures the quality of the classification while significantly reducing the involved processing times.

Spatial-Spectral Feature Extraction With Local Covariance Matrix From Hyperspectral Images Through Hybrid Parallelization

Torti, E;Marenzi, E;Danese, G;Leporati, F
2023-01-01

Abstract

This article presents the optimization and hybrid parallelization of a spatial-spectral feature extraction (FE) method from hyperspectral images (HSIs) using local covariance matrix (CM) representation, exploiting hybrid parallelism through multicore and manycore processors. The aim is to evaluate the performance of parallel versions of this innovative algorithm that characterizes spatial-spectral information prior to classification when conducting FE. The HSI is first projected into a subspace, using the maximum noise fraction method. Then, for each test pixel, its most similar neighbors are clustered using the cosine distance measurement. The result is used to calculate a local CM with each nondiagonal entry characterizing the correlation between different spectral bands. Such matrices represent the spatial-spectral features and are fed to a support vector machine for classification. To optimize the successive parallelization process, a new version of the original MATLAB code has been first developed using C language. This serial version serves as baseline for hybrid parallelization in OpenMP and CUDA. Performance analysis has been conducted using publicly available HSI datasets, confirming that our parallel implementation ensures the quality of the classification while significantly reducing the involved processing times.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11571/1482957
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? 1
social impact