35843ec1525515c530e71654914e51a2 RemSen_a2023v15n18p4422.pdf a817be2aa152103a815c82df5007c633ce99827b RemSen_a2023v15n18p4422.pdf dba11193127907dba23448faf0ac5e3dbfa30b76c6eb425b062c14b54ae6806d RemSen_a2023v15n18p4422.pdf Title: A Scalable Reduced-Complexity Compression of Hyperspectral Remote Sensing Images Using Deep Learning Subject: Two key hurdles to the adoption of Machine Learning (ML) techniques in hyperspectral data compression are computational complexity and scalability for large numbers of bands. These are due to the limited computing capacity available in remote sensing platforms and the high computational cost of compression algorithms for hyperspectral data, especially when the number of bands is large. To address these issues, a channel clusterisation strategy is proposed, which reduces the computational demands of learned compression methods for real scenarios and is scalable for different sources of data with varying numbers of bands. The proposed method is compatible with an embedded implementation for state-of-the-art on board hardware, a first for a ML hyperspectral data compression method. In terms of coding performance, our proposal surpasses established lossy methods such as JPEG 2000 preceded by a spectral Karhunen-Loève Transform (KLT), in clusters of 3 to 7 bands, achieving a PSNR improvement of, on average, 9 dB for AVIRIS and 3 dB for Hyperion images. Keywords: image compression; hyperspectral; deep learning; data compression; AVIRIS; Hyperion Author: Sebastià Mijares i Verdú, Johannes Ballé, Valero Laparra, Joan Bartrina-Rapesta, Miguel Hernández-Cabronero and Joan Serra-Sagristà Creator: LaTeX with hyperref Producer: pdfTeX-1.40.21 CreationDate: Fri Sep 8 10:53:42 2023 CEST ModDate: Fri Sep 8 11:00:35 2023 CEST Custom Metadata: no Metadata Stream: no Tagged: no UserProperties: no Suspects: no Form: none JavaScript: no Pages: 15 Encrypted: no Page size: 595.276 x 841.89 pts (A4) Page rot: 0 File size: 3618994 bytes Optimized: no PDF version: 1.7 name type encoding emb sub uni object ID ------------------------------------ ----------------- ---------------- --- --- --- --------- BKJTQW+URWPalladioL-Bold Type 1 Custom yes yes no 10 0 DBLXVY+URWPalladioL-Roma Type 1 Custom yes yes no 15 0 MWFIWJ+URWPalladioL-Ital Type 1 Custom yes yes no 19 0 ZKJPJD+TimesNewRomanPS-BoldItalicMT Type 1C WinAnsi yes yes no 37 0 QHARRY+CMEX10 Type 1 Builtin yes yes no 62 0 ZASAYP+PazoMath-Italic Type 1 Builtin yes yes no 66 0 JZKDWD+CMR10 Type 1 Builtin yes yes no 70 0 TLCVUG+CMSY10 Type 1 Builtin yes yes no 74 0 RAZVEJ+MSBM10 Type 1 Builtin yes yes no 78 0 QQXUPP+PazoMath Type 1 Builtin yes yes no 110 0 Jhove (Rel. 1.28.0, 2023-05-18) Date: 2024-04-24 02:46:00 CEST RepresentationInformation: RemSen_a2023v15n18p4422.pdf ReportingModule: PDF-hul, Rel. 1.12.4 (2023-03-16) LastModified: 2024-04-23 16:22:16 CEST Size: 3618994 Format: PDF Version: 1.7 Status: Well-Formed and valid SignatureMatches: PDF-hul MIMEtype: application/pdf PDFMetadata: Objects: 295 FreeObjects: 1 IncrementalUpdates: 0 DocumentCatalog: PageLayout: SinglePage PageMode: UseNone Outlines: Item: Title: Introduction Destination: section.1 Item: Title: End-to-End Optimised Transform Coding Destination: section.2 Item: Title: Proposed Method Destination: section.3 Children: Item: Title: Range-Adaptive Normalisation Layer Destination: subsection.3.1 Item: Title: Complexity Reduction Destination: subsection.3.2 Item: Title: Experimental Results Destination: section.4 Item: Title: Conclusions Destination: section.5 Item: Title: References Destination: section.6 Info: Title: A Scalable Reduced-Complexity Compression of Hyperspectral Remote Sensing Images Using Deep Learning Author: Sebastià Mijares i Verdú, Johannes Ballé, Valero Laparra, Joan Bartrina-Rapesta, Miguel Hernández-Cabronero and Joan Serra-Sagristà Subject: Two key hurdles to the adoption of Machine Learning (ML) techniques in hyperspectral data compression are computational complexity and scalability for large numbers of bands. These are due to the limited computing capacity available in remote sensing platforms and the high computational cost of compression algorithms for hyperspectral data, especially when the number of bands is large. To address these issues, a channel clusterisation strategy is proposed, which reduces the computational demands of learned compression methods for real scenarios and is scalable for different sources of data with varying numbers of bands. The proposed method is compatible with an embedded implementation for state-of-the-art on board hardware, a first for a ML hyperspectral data compression method. In terms of coding performance, our proposal surpasses established lossy methods such as JPEG 2000 preceded by a spectral Karhunen-Loève Transform (KLT), in clusters of 3 to 7 bands, achieving a PSNR improvement of, on average, 9 dB for AVIRIS and 3 dB for Hyperion images. Keywords: image compression; hyperspectral; deep learning; data compression; AVIRIS; Hyperion Creator: LaTeX with hyperref Producer: pdfTeX-1.40.21 CreationDate: Fri Sep 08 10:53:42 CEST 2023 ModDate: Fri Sep 08 11:00:35 CEST 2023 ID: 0x048c0c5bd00d6a659812d95e09e735ba, 0x048c0c5bd00d6a659812d95e09e735ba Filters: FilterPipeline: FlateDecode Images: Image: NisoImageMetadata: FormatName: image/png ImageWidth: 1920 ImageHeight: 1080 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Image: NisoImageMetadata: FormatName: image/png ImageWidth: 1920 ImageHeight: 1080 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Image: NisoImageMetadata: FormatName: image/png ImageWidth: 967 ImageHeight: 1080 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Image: NisoImageMetadata: FormatName: image/png ImageWidth: 1920 ImageHeight: 1080 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Image: NisoImageMetadata: FormatName: image/png ImageWidth: 1920 ImageHeight: 1080 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Image: NisoImageMetadata: FormatName: image/png ImageWidth: 1920 ImageHeight: 1080 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Image: NisoImageMetadata: FormatName: image/png ImageWidth: 1074 ImageHeight: 1319 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Image: NisoImageMetadata: FormatName: image/png ImageWidth: 1072 ImageHeight: 1327 ColorSpace: RGB BitsPerSample: 8 BitsPerSampleUnit: integer Filter: FlateDecode Fonts: Type1: Font: BaseFont: ZASAYP+PazoMath-Italic FontSubset: true FirstChar: 35 LastChar: 115 FontDescriptor: FontName: ZASAYP+PazoMath-Italic Flags: Symbolic FontBBox: -70, -277, 902, 733 FontFile: true Font: BaseFont: MWFIWJ+URWPalladioL-Ital FontSubset: true FirstChar: 2 LastChar: 122 FontDescriptor: FontName: MWFIWJ+URWPalladioL-Ital Flags: Symbolic FontBBox: -170, -305, 1010, 941 FontFile: true EncodingDictionary: Differences: true Font: BaseFont: JZKDWD+CMR10 FontSubset: true FirstChar: 40 LastChar: 93 FontDescriptor: FontName: JZKDWD+CMR10 Flags: Symbolic FontBBox: -40, -250, 1009, 750 FontFile: true Font: BaseFont: BKJTQW+URWPalladioL-Bold FontSubset: true FirstChar: 3 LastChar: 250 FontDescriptor: FontName: BKJTQW+URWPalladioL-Bold Flags: Symbolic FontBBox: -152, -301, 1000, 935 FontFile: true EncodingDictionary: Differences: true Font: BaseFont: TLCVUG+CMSY10 FontSubset: true FirstChar: 0 LastChar: 112 FontDescriptor: FontName: TLCVUG+CMSY10 Flags: Symbolic FontBBox: -29, -960, 1116, 775 FontFile: true Font: BaseFont: RAZVEJ+MSBM10 FontSubset: true FirstChar: 69 LastChar: 69 FontDescriptor: FontName: RAZVEJ+MSBM10 Flags: Symbolic FontBBox: -55, -420, 2343, 920 FontFile: true Font: BaseFont: QHARRY+CMEX10 FontSubset: true FirstChar: 0 LastChar: 19 FontDescriptor: FontName: QHARRY+CMEX10 Flags: Symbolic FontBBox: -24, -2960, 1454, 772 FontFile: true Font: BaseFont: QQXUPP+PazoMath FontSubset: true FirstChar: 229 LastChar: 229 FontDescriptor: FontName: QQXUPP+PazoMath Flags: Symbolic FontBBox: -40, -283, 878, 946 FontFile: true Font: BaseFont: DBLXVY+URWPalladioL-Roma FontSubset: true FirstChar: 2 LastChar: 250 FontDescriptor: FontName: DBLXVY+URWPalladioL-Roma Flags: Symbolic FontBBox: -166, -283, 1021, 943 FontFile: true EncodingDictionary: Differences: true Pages: Page: Sequence: 1 Annotations: Annotation: Subtype: Link Rect: 99, 281, 152, 290 Annotation: Subtype: Link Rect: 34, 270, 96, 278 Annotation: Subtype: Link Rect: 34, 164, 96, 186 Annotation: Subtype: Link Rect: 126, 90, 154, 99 Annotation: Subtype: Link Rect: 34, 78, 147, 87 Annotation: Subtype: Link Rect: 34, 756, 208, 792 Annotation: Subtype: Link Rect: 515, 756, 560, 792 Annotation: Subtype: Link Rect: 159, 660, 170, 674 Annotation: Subtype: Link Rect: 247, 660, 258, 674 Annotation: Subtype: Link Rect: 335, 660, 346, 674 Annotation: Subtype: Link Rect: 455, 660, 466, 674 Annotation: Subtype: Link Rect: 179, 648, 190, 662 Annotation: Subtype: Link Rect: 303, 648, 314, 662 Annotation: Subtype: Link Rect: 497, 268, 504, 277 ActionDest: 134 Annotation: Subtype: Link Rect: 192, 231, 199, 240 ActionDest: 134 Annotation: Subtype: Link Rect: 297, 105, 304, 114 ActionDest: 134 Annotation: Subtype: Link Rect: 131, 17, 262, 28 Annotation: Subtype: Link Rect: 385, 17, 560, 28 Annotation: Subtype: Link Rect: 34, 378, 87, 397 Page: Sequence: 2 Annotations: Annotation: Subtype: Link Rect: 351, 723, 358, 732 ActionDest: 134 Annotation: Subtype: Link Rect: 361, 723, 368, 732 ActionDest: 134 Annotation: Subtype: Link Rect: 543, 723, 555, 732 ActionDest: 134 Annotation: Subtype: Link Rect: 218, 710, 230, 719 ActionDest: 134 Annotation: Subtype: Link Rect: 348, 710, 360, 719 ActionDest: 134 Annotation: Subtype: Link Rect: 258, 698, 269, 707 ActionDest: 134 Annotation: Subtype: Link Rect: 272, 698, 284, 707 ActionDest: 134 Annotation: Subtype: Link Rect: 542, 698, 549, 707 ActionDest: 134 Annotation: Subtype: Link Rect: 550, 698, 557, 707 ActionDest: 134 Annotation: Subtype: Link Rect: 221, 660, 233, 669 ActionDest: 134 Annotation: Subtype: Link Rect: 234, 660, 246, 669 ActionDest: 134 Annotation: Subtype: Link Rect: 246, 660, 258, 669 ActionDest: 134 Annotation: Subtype: Link Rect: 203, 647, 215, 657 ActionDest: 134 Annotation: Subtype: Link Rect: 218, 647, 230, 657 ActionDest: 134 Annotation: Subtype: Link Rect: 512, 647, 524, 657 ActionDest: 134 Annotation: Subtype: Link Rect: 524, 648, 536, 657 ActionDest: 134 Annotation: Subtype: Link Rect: 280, 597, 292, 606 ActionDest: 134 Annotation: Subtype: Link Rect: 489, 597, 496, 606 ActionDest: 134 Annotation: Subtype: Link Rect: 336, 585, 348, 594 ActionDest: 134 Annotation: Subtype: Link Rect: 365, 547, 377, 556 ActionDest: 134 Annotation: Subtype: Link Rect: 266, 535, 278, 544 ActionDest: 137 Annotation: Subtype: Link Rect: 423, 510, 435, 519 ActionDest: 137 Annotation: Subtype: Link Rect: 295, 497, 307, 506 ActionDest: 137 Annotation: Subtype: Link Rect: 348, 484, 360, 493 ActionDest: 134 Annotation: Subtype: Link Rect: 361, 484, 373, 493 ActionDest: 134 Annotation: Subtype: Link Rect: 442, 472, 454, 481 ActionDest: 134 Annotation: Subtype: Link Rect: 454, 472, 466, 481 ActionDest: 134 Annotation: Subtype: Link Rect: 467, 472, 478, 481 ActionDest: 137 Annotation: Subtype: Link Rect: 297, 409, 304, 418 ActionDest: 82 Annotation: Subtype: Link Rect: 427, 246, 434, 255 ActionDest: 134 Annotation: Subtype: Link Rect: 543, 183, 555, 192 ActionDest: 137 Page: Sequence: 3 Annotations: Annotation: Subtype: Link Rect: 432, 733, 439, 745 ActionDest: 59 Annotation: Subtype: Link Rect: 475, 720, 482, 732 ActionDest: 82 Annotation: Subtype: Link Rect: 399, 708, 406, 720 ActionDest: 104 Annotation: Subtype: Link Rect: 234, 695, 241, 707 ActionDest: 131 Annotation: Subtype: Link Rect: 242, 560, 249, 568 ActionDest: 134 Annotation: Subtype: Link Rect: 549, 428, 556, 437 ActionDest: 134 Annotation: Subtype: Link Rect: 490, 415, 497, 424 ActionDest: 134 Annotation: Subtype: Link Rect: 191, 288, 198, 299 ActionDest: 59 Annotation: Subtype: Link Rect: 371, 176, 378, 185 ActionDest: 134 Page: Sequence: 4 Annotations: Annotation: Subtype: Link Rect: 512, 719, 524, 728 ActionDest: 137 Annotation: Subtype: Link Rect: 389, 694, 396, 703 ActionDest: 134 Annotation: Subtype: Link Rect: 465, 656, 472, 665 ActionDest: 134 Annotation: Subtype: Link Rect: 552, 654, 559, 666 ActionDest: 82 Annotation: Subtype: Link Rect: 336, 581, 348, 590 ActionDest: 134 Annotation: Subtype: Link Rect: 249, 443, 263, 452 ActionDest: 93 Annotation: Subtype: Link Rect: 279, 206, 286, 214 ActionDest: 134 Annotation: Subtype: Link Rect: 529, 159, 541, 168 ActionDest: 137 Page: Sequence: 5 Annotations: Annotation: Subtype: Link Rect: 447, 720, 454, 732 ActionDest: 82 Annotation: Subtype: Link Rect: 509, 673, 521, 682 ActionDest: 137 Annotation: Subtype: Link Rect: 430, 660, 442, 669 ActionDest: 137 Annotation: Subtype: Link Rect: 544, 660, 556, 669 ActionDest: 137 Annotation: Subtype: Link Rect: 548, 633, 555, 644 ActionDest: 82 Annotation: Subtype: Link Rect: 385, 483, 392, 492 ActionDest: 134 Annotation: Subtype: Link Rect: 433, 480, 440, 492 ActionDest: 88 Annotation: Subtype: Link Rect: 321, 470, 328, 479 ActionDest: 134 Annotation: Subtype: Link Rect: 312, 379, 319, 391 ActionDest: 93 Page: Sequence: 6 Annotations: Annotation: Subtype: Link Rect: 543, 348, 555, 357 ActionDest: 134 Annotation: Subtype: Link Rect: 279, 320, 286, 332 ActionDest: 82 Annotation: Subtype: Link Rect: 204, 310, 216, 319 ActionDest: 134 Annotation: Subtype: Link Rect: 212, 270, 219, 282 ActionDest: 98 Annotation: Subtype: Link Rect: 207, 232, 214, 244 ActionDest: 82 Annotation: Subtype: Link Rect: 228, 173, 235, 185 ActionDest: 93 Page: Sequence: 7 Annotations: Annotation: Subtype: Link Rect: 212, 401, 219, 413 ActionDest: 101 Annotation: Subtype: Link Rect: 548, 339, 555, 350 ActionDest: 98 Annotation: Subtype: Link Rect: 544, 203, 556, 212 ActionDest: 137 Annotation: Subtype: Link Rect: 197, 153, 209, 162 ActionDest: 137 Annotation: Subtype: Link Rect: 210, 153, 221, 162 ActionDest: 137 Annotation: Subtype: Link Rect: 325, 128, 337, 137 ActionDest: 134 Page: Sequence: 8 Annotations: Annotation: Subtype: Link Rect: 494, 651, 506, 660 ActionDest: 134 Annotation: Subtype: Link Rect: 489, 636, 496, 648 ActionDest: 101 Annotation: Subtype: Link Rect: 458, 510, 465, 522 ActionDest: 104 Annotation: Subtype: Link Rect: 370, 476, 381, 484 ActionDest: 134 Page: Sequence: 9 Annotations: Annotation: Subtype: Link Rect: 367, 736, 378, 745 ActionDest: 134 Annotation: Subtype: Link Rect: 325, 385, 337, 394 ActionDest: 134 Annotation: Subtype: Link Rect: 424, 385, 436, 394 ActionDest: 137 Annotation: Subtype: Link Rect: 544, 385, 556, 394 ActionDest: 137 Annotation: Subtype: Link Rect: 398, 347, 410, 356 ActionDest: 137 Annotation: Subtype: Link Rect: 413, 347, 425, 356 ActionDest: 137 Annotation: Subtype: Link Rect: 458, 310, 470, 319 ActionDest: 134 Annotation: Subtype: Link Rect: 344, 260, 356, 269 ActionDest: 137 Annotation: Subtype: Link Rect: 544, 209, 556, 218 ActionDest: 137 Annotation: Subtype: Link Rect: 306, 197, 318, 206 ActionDest: 137 Annotation: Subtype: Link Rect: 544, 197, 556, 206 ActionDest: 137 Annotation: Subtype: Link Rect: 226, 147, 238, 156 ActionDest: 137 Annotation: Subtype: Link Rect: 465, 81, 560, 93 Annotation: Subtype: Link Rect: 165, 69, 260, 81 Page: Sequence: 10 Annotations: Annotation: Subtype: Link Rect: 217, 733, 224, 745 ActionDest: 107 Annotation: Subtype: Link Rect: 241, 695, 248, 707 ActionDest: 107 Annotation: Subtype: Link Rect: 451, 120, 458, 132 ActionDest: 118 Annotation: Subtype: Link Rect: 423, 107, 430, 119 ActionDest: 107 Annotation: Subtype: Link Rect: 467, 107, 474, 119 ActionDest: 118 Annotation: Subtype: Link Rect: 493, 107, 499, 119 ActionDest: 126 Page: Sequence: 11 Page: Sequence: 12 Annotations: Annotation: Subtype: Link Rect: 204, 199, 216, 208 ActionDest: 137 Annotation: Subtype: Link Rect: 464, 197, 470, 208 ActionDest: 131 Annotation: Subtype: Link Rect: 202, 184, 209, 196 ActionDest: 131 Page: Sequence: 13 Annotations: Annotation: Subtype: Link Rect: 247, 683, 258, 692 ActionDest: 137 Annotation: Subtype: Link Rect: 204, 128, 251, 139 Page: Sequence: 14 Annotations: Annotation: Subtype: Link Rect: 171, 590, 208, 600 Annotation: Subtype: Link Rect: 221, 578, 440, 589 Annotation: Subtype: Link Rect: 282, 543, 320, 554 Annotation: Subtype: Link Rect: 252, 451, 289, 462 Annotation: Subtype: Link Rect: 59, 291, 96, 301 Annotation: Subtype: Link Rect: 108, 268, 146, 278 Annotation: Subtype: Link Rect: 333, 244, 371, 255 Annotation: Subtype: Link Rect: 190, 222, 228, 232 Annotation: Subtype: Link Rect: 139, 199, 177, 209 Annotation: Subtype: Link Rect: 305, 175, 342, 186 Annotation: Subtype: Link Rect: 341, 152, 378, 163 Annotation: Subtype: Link Rect: 139, 95, 177, 105 Annotation: Subtype: Link Rect: 309, 72, 347, 82 Page: Sequence: 15 Annotations: Annotation: Subtype: Link Rect: 376, 722, 414, 733 Annotation: Subtype: Link Rect: 214, 699, 251, 710 Annotation: Subtype: Link Rect: 108, 642, 146, 652 Annotation: Subtype: Link Rect: 226, 596, 264, 606 Annotation: Subtype: Link Rect: 449, 584, 486, 594 Annotation: Subtype: Link Rect: 59, 515, 96, 525 Annotation: Subtype: Link Rect: 188, 422, 226, 433 Annotation: Subtype: Link Rect: 219, 400, 257, 410 Annotation: Subtype: Link Rect: 168, 377, 205, 387 Annotation: Subtype: Link Rect: 342, 365, 501, 376 Annotation: Subtype: Link Rect: 215, 342, 340, 353 Annotation: Subtype: Link Rect: 265, 319, 302, 330 Checksum: 6d2f7cf4 Type: CRC32 Checksum: 35843ec1525515c530e71654914e51a2 Type: MD5 Checksum: a817be2aa152103a815c82df5007c633ce99827b Type: SHA-1 Checksum: dba11193127907dba23448faf0ac5e3dbfa30b76c6eb425b062c14b54ae6806d Type: SHA-256