Comparison of Lightweight Deep Neural Networks for Landsat Time-Series Land Use and Land Cover Classification over the Conterminous United States.
Saved in:
| Title: | Comparison of Lightweight Deep Neural Networks for Landsat Time-Series Land Use and Land Cover Classification over the Conterminous United States. |
|---|---|
| Authors: | Wang, Zhixin1 (AUTHOR), Mountrakis, Giorgos1 (AUTHOR) gmountrakis@esf.edu, Safaeinia, Ahmadreza1 (AUTHOR) |
| Source: | Remote Sensing. Jun2026, Vol. 18 Issue 11, p1757. 29p. |
| Subjects: | Land use mapping, Recurrent neural networks, Artificial neural networks, Environmental monitoring |
| Abstract: | Highlights: What are the main findings? Simple Recurrent Unit-based lightweight models consistently outperformed traditional classifiers with small model sizes. From the tested SRU models, MobileNet offered the greatest improvement. What are the implications of the main findings? When model complexity is constrained due to limited labels or computing resources, selecting lightweight models can improve the trade-off between model efficiency and classification accuracy. Accurate and timely land cover and land use (LCLU) classification from medium-spatial-resolution optical time-series data is essential for large-scale environmental monitoring. lightweight deep neural networks (DNNs) offer reduced computational and memory requirements, enabling efficient deployment in resource-constrained scenarios. While popular in computer vision tasks, their ability to simultaneously model spatial, spectral, and temporal information for medium-resolution optical time series is understudied. This study addresses this gap by evaluating seven existing lightweight models spanning four architectural families: convolutional and recurrent hybrids, convolutional and transformer hybrids, 3D convolutional models, and video transformers against a traditional hybrid convolutional transformer (CNNTransformer) benchmark across the Conterminous United States (CONUS). Models are trained on 500,000 Landsat time-series samples with 25 repetitions and evaluated across five model sizes (3k, 5k, 10k, 25k, and 50k parameters) to assess both accuracy and stability. Results show that Simple Recurrent Unit (SRU)-based lightweight hybrids provide the best performance. Specifically, MobileNetSRU consistently outperformed the benchmark at small-to-moderate model sizes (3k–15k), achieving peak relative improvement gains of ~2.5–7.5% at 7.5k parameters. MobileNetSRU also demonstrated superior robustness in limited-data scenarios (50k training samples), particularly for spectrally stable classes like water and bare land. This study reveals that the inherent inductive bias of recurrent-based lightweight models aligns more effectively with the sequential phenology of satellite data than more flexible, data-hungry attention mechanisms at small parameter scales. These findings suggest that strategically matching architectural priorities to temporal data structures can significantly reduce the trade-off between model efficiency and classification accuracy in scalable Earth-observation workflows. [ABSTRACT FROM AUTHOR] |
| Copyright of Remote Sensing is the property of MDPI and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| FullText | Links: – Type: pdflink Text: Availability: 1 |
|---|---|
| Header | DbId: egs DbLabel: Engineering Source An: 194586978 AccessLevel: 6 PubType: Academic Journal PubTypeId: academicJournal PreciseRelevancyScore: 0 |
| IllustrationInfo | |
| Items | – Name: Title Label: Title Group: Ti Data: Comparison of Lightweight Deep Neural Networks for Landsat Time-Series Land Use and Land Cover Classification over the Conterminous United States. – Name: Author Label: Authors Group: Au Data: <searchLink fieldCode="AR" term="%22Wang%2C+Zhixin%22">Wang, Zhixin</searchLink><relatesTo>1</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Mountrakis%2C+Giorgos%22">Mountrakis, Giorgos</searchLink><relatesTo>1</relatesTo> (AUTHOR)<i> gmountrakis@esf.edu</i><br /><searchLink fieldCode="AR" term="%22Safaeinia%2C+Ahmadreza%22">Safaeinia, Ahmadreza</searchLink><relatesTo>1</relatesTo> (AUTHOR) – Name: TitleSource Label: Source Group: Src Data: <searchLink fieldCode="JN" term="%22Remote+Sensing%22">Remote Sensing</searchLink>. Jun2026, Vol. 18 Issue 11, p1757. 29p. – Name: Subject Label: Subjects Group: Su Data: <searchLink fieldCode="DE" term="%22Land+use+mapping%22">Land use mapping</searchLink><br /><searchLink fieldCode="DE" term="%22Recurrent+neural+networks%22">Recurrent neural networks</searchLink><br /><searchLink fieldCode="DE" term="%22Artificial+neural+networks%22">Artificial neural networks</searchLink><br /><searchLink fieldCode="DE" term="%22Environmental+monitoring%22">Environmental monitoring</searchLink> – Name: Abstract Label: Abstract Group: Ab Data: Highlights: What are the main findings? Simple Recurrent Unit-based lightweight models consistently outperformed traditional classifiers with small model sizes. From the tested SRU models, MobileNet offered the greatest improvement. What are the implications of the main findings? When model complexity is constrained due to limited labels or computing resources, selecting lightweight models can improve the trade-off between model efficiency and classification accuracy. Accurate and timely land cover and land use (LCLU) classification from medium-spatial-resolution optical time-series data is essential for large-scale environmental monitoring. lightweight deep neural networks (DNNs) offer reduced computational and memory requirements, enabling efficient deployment in resource-constrained scenarios. While popular in computer vision tasks, their ability to simultaneously model spatial, spectral, and temporal information for medium-resolution optical time series is understudied. This study addresses this gap by evaluating seven existing lightweight models spanning four architectural families: convolutional and recurrent hybrids, convolutional and transformer hybrids, 3D convolutional models, and video transformers against a traditional hybrid convolutional transformer (CNNTransformer) benchmark across the Conterminous United States (CONUS). Models are trained on 500,000 Landsat time-series samples with 25 repetitions and evaluated across five model sizes (3k, 5k, 10k, 25k, and 50k parameters) to assess both accuracy and stability. Results show that Simple Recurrent Unit (SRU)-based lightweight hybrids provide the best performance. Specifically, MobileNetSRU consistently outperformed the benchmark at small-to-moderate model sizes (3k–15k), achieving peak relative improvement gains of ~2.5–7.5% at 7.5k parameters. MobileNetSRU also demonstrated superior robustness in limited-data scenarios (50k training samples), particularly for spectrally stable classes like water and bare land. This study reveals that the inherent inductive bias of recurrent-based lightweight models aligns more effectively with the sequential phenology of satellite data than more flexible, data-hungry attention mechanisms at small parameter scales. These findings suggest that strategically matching architectural priorities to temporal data structures can significantly reduce the trade-off between model efficiency and classification accuracy in scalable Earth-observation workflows. [ABSTRACT FROM AUTHOR] – Name: AbstractSuppliedCopyright Label: Group: Ab Data: <i>Copyright of Remote Sensing is the property of MDPI and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract.</i> (Copyright applies to all Abstracts.) |
| PLink | https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=egs&AN=194586978 |
| RecordInfo | BibRecord: BibEntity: Identifiers: – Type: doi Value: 10.3390/rs18111757 Languages: – Code: eng Text: English PhysicalDescription: Pagination: PageCount: 29 StartPage: 1757 Subjects: – SubjectFull: Land use mapping Type: general – SubjectFull: Recurrent neural networks Type: general – SubjectFull: Artificial neural networks Type: general – SubjectFull: Environmental monitoring Type: general Titles: – TitleFull: Comparison of Lightweight Deep Neural Networks for Landsat Time-Series Land Use and Land Cover Classification over the Conterminous United States. Type: main BibRelationships: HasContributorRelationships: – PersonEntity: Name: NameFull: Wang, Zhixin – PersonEntity: Name: NameFull: Mountrakis, Giorgos – PersonEntity: Name: NameFull: Safaeinia, Ahmadreza IsPartOfRelationships: – BibEntity: Dates: – D: 01 M: 06 Text: Jun2026 Type: published Y: 2026 Identifiers: – Type: issn-print Value: 20724292 Numbering: – Type: volume Value: 18 – Type: issue Value: 11 Titles: – TitleFull: Remote Sensing Type: main |
| ResultId | 1 |
