A Novel Image Fusion Method of Multi-Spectral and SAR Images for Land Cover Classification

Abstract

The fusion of multi-spectral and synthetic aperture radar (SAR) images could retain the advantages of each data, hence benefiting accurate land cover classification. However, some current image fusion methods face the challenge of producing unexpected noise. To overcome the aforementioned problem, this paper proposes a novel fusion method based on weighted median filter and Gram–Schmidt transform. In the proposed method, Sentinel-2A images and GF-3 images are respectively subjected to different preprocessing processes. Since weighted median filter does not strongly blur edges while filtering, it is applied to Sentinel-2A images for reducing noise. The processed Sentinel images are then transformed by Gram–Schmidt with GF-3 images. Two popular methods, principal component analysis method and traditional Gram–Schmidt transform, are used as the comparison methods in the experiment. In addition, random forest, a powerful ensemble model, is adopted as the land cover classifier due to its fast training speed and excellent classification performance. The overall accuracy, Kappa coefficient and classification map of the random forest are used as the evaluation criteria of the fusion method. Experiments conducted on five datasets demonstrate the superiority of the proposed method in both objective metrics and visual impressions. The experimental results indicate that the proposed method can improve the overall accuracy by up to 5% compared to using the original Sentinel-2A and has the potential to improve the satellite-based land cover classification accuracy.

Comparison and assessment of different object-based classifications using machine learning algorithms and UAVs multispectral imagery: a case study in a citrus orchard and an onion crop

“…It is a non-parametric test based on a contingency table of 2 × 2 dimension constructed using the binary distinction between correct and incorrect class allocations, widely used to determine if statistically significant differences are present between pairs of classifications (Y. Gao et al, 2011);Millard & Richardson, 2015;Quan et al, 2020;Belgiu & Csillik, 2018;Kavzoglu, 2017;Dietterich, 1998;Sokolova et al, 2006). This test is suitable to assess the difference of multiple classification accuracy performances when the same set of validation and training samples are used (Foody, 2004;Y.…”

Relative Total Variation Structure Analysis-Based Fusion Method for Hyperspectral and LiDAR Data Classification

“…Random forest (RF) is a standard method of ensemble learning, with the outstanding output of classification and high processing speed, and it can prevent over-fitting effectively [51][52][53][54][55][56][57]. RF is a mixture of tree predictors.…”

“…The parameters of the RTVSA method were set as follows: λ = 0.04, σ = 2, and L = 30. In image classification, according to the results in Reference [62] and our previous work [51], the number of decision trees in the RF was determined to be 100, and the number of prediction variables was set approximately to the square root of the number of input bands. The settings of the CNN were based on [59].…”

Entropy Indicators: An Approach for Low-Speed Bearing Diagnosis

“…To analyze how entropy-based and CIs can complement each other, the combination of both types of indicators is studied. As the RF classifier provides excellent performance in pattern recognition [ 66 , 67 , 68 , 69 , 70 ], this study attempts to utilize the RF classifier to construct a classification model for fault recognition of low-speed bearings.…”