Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant Models

Artificial intelligence (AI) has become the mainstream of analysis tools in remote sensing. Various semantic segmentation models have been introduced to segment land cover from aerial or satellite images, and remarkable results have been achieved. However, they often lack universal performance on un...

Full description

Saved in:
Bibliographic Details
Main Authors: Jae Young Chang, Kwan-Young Oh, Kwang-Jae Lee
Format: Article
Language:English
Published: MDPI AG 2025-08-01
Series:Remote Sensing
Subjects:
Online Access:https://www.mdpi.com/2072-4292/17/15/2669
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849405950908170240
author Jae Young Chang
Kwan-Young Oh
Kwang-Jae Lee
author_facet Jae Young Chang
Kwan-Young Oh
Kwang-Jae Lee
author_sort Jae Young Chang
collection DOAJ
description Artificial intelligence (AI) has become the mainstream of analysis tools in remote sensing. Various semantic segmentation models have been introduced to segment land cover from aerial or satellite images, and remarkable results have been achieved. However, they often lack universal performance on unseen images, making them challenging to provide as a service. One of the primary reasons for the lack of robustness is overfitting, resulting from errors and inconsistencies in the ground truth (GT). In this study, we propose a method to mitigate these inconsistencies by utilizing redundant models and verify the improvement using a public dataset based on Google Earth images. Redundant models share the same network architecture and hyperparameters but are trained with different combinations of training and validation data on the same dataset. Because of the variations in sample exposure during training, these models yield slightly different inference results. This variability allows for the estimation of pixel-level confidence levels for the GT. The confidence level is incorporated into the GT to influence the loss calculation during the training of the enhanced model. Furthermore, we implemented a consensus model that employs modified masks, where classes with low confidence are substituted by the dominant classes identified through a majority vote from the redundant models. To further improve robustness, we extended the same approach to fuse the dataset with different class compositions based on imagery from the Korea Multipurpose Satellite 3A (KOMPSAT-3A). Performance evaluations were conducted on three network architectures: a simple network, U-Net, and DeepLabV3. In the single-dataset case, the performance of the enhanced and consensus models improved by an average of 2.49% and 2.59% across the network architectures. In the multi-dataset scenario, the enhanced models and consensus models showed an average performance improvement of 3.37% and 3.02% across the network architectures, respectively, compared to an average increase of 1.55% without the proposed method.
format Article
id doaj-art-609536bcc2ee4a68999f0cf773ac276a
institution Kabale University
issn 2072-4292
language English
publishDate 2025-08-01
publisher MDPI AG
record_format Article
series Remote Sensing
spelling doaj-art-609536bcc2ee4a68999f0cf773ac276a2025-08-20T03:36:32ZengMDPI AGRemote Sensing2072-42922025-08-011715266910.3390/rs17152669Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant ModelsJae Young Chang0Kwan-Young Oh1Kwang-Jae Lee2National Satellite Operation & Application Center, Korea Aerospace Research Institute, Daejeon 34133, Republic of KoreaNational Satellite Operation & Application Center, Korea Aerospace Research Institute, Daejeon 34133, Republic of KoreaNational Satellite Operation & Application Center, Korea Aerospace Research Institute, Daejeon 34133, Republic of KoreaArtificial intelligence (AI) has become the mainstream of analysis tools in remote sensing. Various semantic segmentation models have been introduced to segment land cover from aerial or satellite images, and remarkable results have been achieved. However, they often lack universal performance on unseen images, making them challenging to provide as a service. One of the primary reasons for the lack of robustness is overfitting, resulting from errors and inconsistencies in the ground truth (GT). In this study, we propose a method to mitigate these inconsistencies by utilizing redundant models and verify the improvement using a public dataset based on Google Earth images. Redundant models share the same network architecture and hyperparameters but are trained with different combinations of training and validation data on the same dataset. Because of the variations in sample exposure during training, these models yield slightly different inference results. This variability allows for the estimation of pixel-level confidence levels for the GT. The confidence level is incorporated into the GT to influence the loss calculation during the training of the enhanced model. Furthermore, we implemented a consensus model that employs modified masks, where classes with low confidence are substituted by the dominant classes identified through a majority vote from the redundant models. To further improve robustness, we extended the same approach to fuse the dataset with different class compositions based on imagery from the Korea Multipurpose Satellite 3A (KOMPSAT-3A). Performance evaluations were conducted on three network architectures: a simple network, U-Net, and DeepLabV3. In the single-dataset case, the performance of the enhanced and consensus models improved by an average of 2.49% and 2.59% across the network architectures. In the multi-dataset scenario, the enhanced models and consensus models showed an average performance improvement of 3.37% and 3.02% across the network architectures, respectively, compared to an average increase of 1.55% without the proposed method.https://www.mdpi.com/2072-4292/17/15/2669satellite imageryKOMPSATdeep-learningland cover segmentationoverfittingsemantic segmentation
spellingShingle Jae Young Chang
Kwan-Young Oh
Kwang-Jae Lee
Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant Models
Remote Sensing
satellite imagery
KOMPSAT
deep-learning
land cover segmentation
overfitting
semantic segmentation
title Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant Models
title_full Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant Models
title_fullStr Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant Models
title_full_unstemmed Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant Models
title_short Improving the Universal Performance of Land Cover Semantic Segmentation Through Training Data Refinement and Multi-Dataset Fusion via Redundant Models
title_sort improving the universal performance of land cover semantic segmentation through training data refinement and multi dataset fusion via redundant models
topic satellite imagery
KOMPSAT
deep-learning
land cover segmentation
overfitting
semantic segmentation
url https://www.mdpi.com/2072-4292/17/15/2669
work_keys_str_mv AT jaeyoungchang improvingtheuniversalperformanceoflandcoversemanticsegmentationthroughtrainingdatarefinementandmultidatasetfusionviaredundantmodels
AT kwanyoungoh improvingtheuniversalperformanceoflandcoversemanticsegmentationthroughtrainingdatarefinementandmultidatasetfusionviaredundantmodels
AT kwangjaelee improvingtheuniversalperformanceoflandcoversemanticsegmentationthroughtrainingdatarefinementandmultidatasetfusionviaredundantmodels