Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing Images
We investigated nearest-neighbor density-based clustering for hyperspectral image analysis. Four existing techniques were considered that rely on a K-nearest neighbor (KNN) graph to estimate local density and to propagate labels through algorithm-specific labeling decisions. We first improved two of...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-11-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/12/22/3745 |
| _version_ | 1797547901431316480 |
|---|---|
| author | Claude Cariou Steven Le Moan Kacem Chehdi |
| author_facet | Claude Cariou Steven Le Moan Kacem Chehdi |
| author_sort | Claude Cariou |
| collection | DOAJ |
| description | We investigated nearest-neighbor density-based clustering for hyperspectral image analysis. Four existing techniques were considered that rely on a K-nearest neighbor (KNN) graph to estimate local density and to propagate labels through algorithm-specific labeling decisions. We first improved two of these techniques, a KNN variant of the density peaks clustering method <span style="font-variant: small-caps;">dpc</span>, and a weighted-mode variant of <span style="font-variant: small-caps;">knnclust</span>, so the four methods use the same input KNN graph and only differ by their labeling rules. We propose two regularization schemes for hyperspectral image analysis: (i) a graph regularization based on mutual nearest neighbors (MNN) prior to clustering to improve cluster discovery in high dimensions; (ii) a spatial regularization to account for correlation between neighboring pixels. We demonstrate the relevance of the proposed methods on synthetic data and hyperspectral images, and show they achieve superior overall performances in most cases, outperforming the state-of-the-art methods by up to 20% in kappa index on real hyperspectral images. |
| first_indexed | 2024-03-10T14:51:40Z |
| format | Article |
| id | doaj.art-4625f87c0e3b46ef82aab25f072f3569 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-10T14:51:40Z |
| publishDate | 2020-11-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-4625f87c0e3b46ef82aab25f072f35692023-11-20T20:57:11ZengMDPI AGRemote Sensing2072-42922020-11-011222374510.3390/rs12223745Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing ImagesClaude Cariou0Steven Le Moan1Kacem Chehdi2Institut d’Électronique et des Technologies du numéRique, CNRS, Univ Rennes, UMR 6164, Enssat, 6 rue de Kerampont, F22300 Lannion, FranceCentre for Research in Image and Signal Processing, Massey University, Palmerston North 4410, New ZealandInstitut d’Électronique et des Technologies du numéRique, CNRS, Univ Rennes, UMR 6164, Enssat, 6 rue de Kerampont, F22300 Lannion, FranceWe investigated nearest-neighbor density-based clustering for hyperspectral image analysis. Four existing techniques were considered that rely on a K-nearest neighbor (KNN) graph to estimate local density and to propagate labels through algorithm-specific labeling decisions. We first improved two of these techniques, a KNN variant of the density peaks clustering method <span style="font-variant: small-caps;">dpc</span>, and a weighted-mode variant of <span style="font-variant: small-caps;">knnclust</span>, so the four methods use the same input KNN graph and only differ by their labeling rules. We propose two regularization schemes for hyperspectral image analysis: (i) a graph regularization based on mutual nearest neighbors (MNN) prior to clustering to improve cluster discovery in high dimensions; (ii) a spatial regularization to account for correlation between neighboring pixels. We demonstrate the relevance of the proposed methods on synthetic data and hyperspectral images, and show they achieve superior overall performances in most cases, outperforming the state-of-the-art methods by up to 20% in kappa index on real hyperspectral images.https://www.mdpi.com/2072-4292/12/22/3745clustering methodsdensity estimationnearest neighbor searchdeterministic algorithmunsupervised learning |
| spellingShingle | Claude Cariou Steven Le Moan Kacem Chehdi Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing Images Remote Sensing clustering methods density estimation nearest neighbor search deterministic algorithm unsupervised learning |
| title | Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing Images |
| title_full | Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing Images |
| title_fullStr | Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing Images |
| title_full_unstemmed | Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing Images |
| title_short | Improving K-Nearest Neighbor Approaches for Density-Based Pixel Clustering in Hyperspectral Remote Sensing Images |
| title_sort | improving k nearest neighbor approaches for density based pixel clustering in hyperspectral remote sensing images |
| topic | clustering methods density estimation nearest neighbor search deterministic algorithm unsupervised learning |
| url | https://www.mdpi.com/2072-4292/12/22/3745 |
| work_keys_str_mv | AT claudecariou improvingknearestneighborapproachesfordensitybasedpixelclusteringinhyperspectralremotesensingimages AT stevenlemoan improvingknearestneighborapproachesfordensitybasedpixelclusteringinhyperspectralremotesensingimages AT kacemchehdi improvingknearestneighborapproachesfordensitybasedpixelclusteringinhyperspectralremotesensingimages |