Comparative Analysis of Different Image Fusion Methods at the Pixel and Decision Levels on the Accuracy of Land Use and Land Cover Classification

Abstract

Introduction: This research examines and evaluates various methods of fusion satellite images to produce high accuracy land use and land cover maps in the Ahvaz region. Considering the importance of accurate land use information in natural resource management, urban planning, and sustainable development, this study aims to analyze different integration methods and investigate their impact on the accuracy of land use classification. In this context, two major levels of data integration are explored: pixel-level integration and decision-level integration. Pixel-level integration involves combining information from multiple images simultaneously at the individual pixel level, which can significantly improve the accuracy and quality of the final images. On the other hand, decision-level integration focuses on combining the results obtained from different classification algorithms.

Materials and Methods: In this research, images from two sensors, Landsat 8 and Sentinel 2, were used. Landsat 8, with a spatial resolution of 30 meters, and Sentinel 2, with a spatial resolution of 10 meters, were specifically chosen for applications related to land use and land cover.In the first step, images were integrated at the pixel level using various integration methods, including Discrete Wavelet Transform (DWT), Spatial Filtering based on Intensity Modulation (SFIM), Gram Schmidt (GS), Multiplicative (MP), Brovey Transform and Principal Component Analysis (PC). Each of these methods was specifically designed to preserve the spectral and spatial characteristics in the integrated images.In the second step, two classification methods, Maximum Likelihood Classification (MLC) and Support Vector Machine (SVM), were employed to create classified images. This choice was made because of the high ability of both methods to differentiate various land use and land cover classes. As a final step, the use of the Dempster-Shafer method as a decision-level integration approach was examined. This method allows for the combination of evidence and information from various sources, facilitating the production of more accurate and reliable results.

Results and discussion: The results of this study indicated that the Support Vector Machine (SVM) method achieved higher accuracy in land use and land cover classification compared to the Maximum Likelihood Classification (MLC) method. The land use classification derived from the SFIM image using MLC, along with those obtained from GS, PC, and Brovey using SVM, were selected for decision-level integration due to their high Kappa coefficient and overall accuracy. The final land use map obtained through Dempster-Shafer integration exhibited an overall accuracy of 98.38% and a Kappa coefficient of 97.67%. These results reflect an improvement of 5 to 7 percent compared to the four land uses utilized. The increase in accuracy in class differentiation signifies success in more precise identification and classification of land use classes. Furthermore, the results revealed that the Dempster-Shafer method provided significant improvements, particularly in distinguishing similar classes such as soil and residential-road, leading to a notable increase in producer accuracy for these classes.Additionally, the examination of confusion matrices showed that the application of the Dempster-Shafer method reduced ambiguity in the classification of various land use classes. This emphasizes that the correct selection of integration and classification methods can directly enhance the accuracy and quality of land use maps.

Conclusion: This research clearly demonstrates that the use of the Dempster-Shafer method as an effective tool in integrating classified data can significantly increase the accuracy of land use maps. Moreover, this study emphasizes that selecting appropriate integration methods at both the pixel and decision levels can positively impact the quality of land use maps. Ultimately, this research serves as a valuable reference for future studies in the field of remote sensing data integration and its applications in natural resource management and urban planning, underscoring the importance of having accurate land use maps.