Fusion of SAR and Multispectral Images Using Random Forest Regression for Change Detection

Seo, Dae Kyo; Kim, Yong Hyun; Eo, Yang Dam; Lee, Mi Hee; Park, Wan Yong

doi:10.3390/ijgi7100401

Cited by 51 publications

(26 citation statements)

References 38 publications

(58 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A single pixel does not contain enough information and, therefore, features other than the pixel value must be considered to perform the phenological normalization [28,38,39]. The spectral indices are considered as features in this study, which is influenced by phenological properties [39][40][41].…”

Section: Selection Of the Spectral Indexmentioning

confidence: 99%

“…Changes that are caused by vegetation in particular have the most typical characteristics, including nonlinearity, which induces serious disturbances when change detection is performed [25,26]. In addition, since optical satellite imagery, which is the main source in change detection, is affected by clouds and atmospheric conditions, it is difficult to acquire images that meet the temporal requirements [27,28]. In such cases, images with phenological differences as well as radiometric differences should be utilized [25,26].…”

mentioning

confidence: 99%

See 1 more Smart Citation

Multilayer Perceptron-Based Phenological and Radiometric Normalization for High-Resolution Satellite Imagery

Seo

2019

Applied Sciences

Self Cite

View full text Add to dashboard Cite

Radiometric normalization is an essential preprocessing step that must be performed to detect changes in multi-temporal satellite images and, in general, relative radiometric normalization is utilized. However, most relative radiometric normalization methods assume a linear relationship and they cannot take into account nonlinear properties, such as the distribution of the earth's surface or phenological differences that are caused by the growth of vegetation. Thus, this paper proposes a novel method that assumes a nonlinear relationship and it uses a representative nonlinear regression model-multilayer perceptron (MLP). The proposed method performs radiometric resolution compression while considering both the complexity and time cost, and radiometric control set samples are extracted based on a no-change set method. Subsequently, the spectral index is selected for each band to compensate for the phenological properties, phenological normalization is performed based on MLP, and the global radiometric properties are adjusted through postprocessing. Finally, a performance evaluation is conducted by comparing the results herein with those from conventional relative radiometric normalization algorithms. The experimental results show that the proposed method outperforms conventional methods in terms of both visual inspection and quantitative evaluation. In other words, the applicability of the proposed method to the normalization of multi-temporal images with nonlinear properties is confirmed. Relative radiometric normalization, on the other hand, is an image-based normalization method that selects one image as a reference and matches the radiometric characteristics of a subject image [13,14]. That is, relative radiometric normalization is generally utilized since the multi-temporal images are normalized into common scales without extra parameters [12,15].Relative radiometric normalization can be classified into two categories: a global statistics-based method and a radiometric control set sample (RCSS)-based method. The former is performed by using statistical values of all pixels in the image, which includes histogram matching, minimum-maximum (MM), mean-standard deviation (MS), and simple regression (SR) [15,16]. In contrast, the latter is based on invariant pixels, known as RCSSs, which include dark set-bright set (DB), pseudo invariant features (PIF), and no-change sets (NC) [17][18][19][20]. At this time, except for histogram matching, normalization is performed through regression equations, while assuming a linear relationship between the pixels at the same position in each band [21]. However, most remote sensing data are nonlinearly distributed, and the surface of the earth is composed of a complex mixture of natural and man-made features that exhibit nonlinear characteristics [22][23][24]. Changes that are caused by vegetation in particular have the most typical characteristics, including nonlinearity, which induces serious disturbances when change detection is performed [25,26]. In addition, since optical sa...

show abstract

Section: Selection Of the Spectral Indexmentioning

confidence: 99%

mentioning

confidence: 99%

Multilayer Perceptron-Based Phenological and Radiometric Normalization for High-Resolution Satellite Imagery

Seo

2019

Applied Sciences

Self Cite

View full text Add to dashboard Cite

show abstract

“…• Random Forest Regression [29]: applied individually to a RGB multispectral image and to a SAR image pre-processed by gray-level co-occurrence matrix descriptors (e.g., energy, homogeneity, angular second moment), to train independent classes by K-means, then training samples are selected and used in the learning process, determining the RF for each class, then applying the resulting RF to the entire image. This results in the fusion of thesurface roughness characteristics of the SAR image and the spectral characteristics of the MS image; • Two-Branch Convolutional Neural Network [30]: CNN layers are applied to the respective input features: a principal component analysis (PCA) is performed on an hyper-spectral 144-bands image (branch 1-image provided at the IEEE-GRSS 2013 data fusion contest), then, in next layers, a series of filters whose parameters are tuned by supervised learning.…”

Section: What Does This Special Issue Bring To the Reader?mentioning

confidence: 99%

Multi-Source Geo-Information Fusion in Transition: A Summer 2019 Snapshot

Jeansoulin¹

2019

IJGI

View full text Add to dashboard Cite

Since the launch of Landsat-1 in 1972, the scientific domain of geo-information has been incrementally shaped through different periods, due to technology evolutions: in devices (satellites, UAV, IoT), in sensors (optical, radar, LiDAR), in software (GIS, WebGIS, 3D), and in communication (Big Data). Land Cover and Disaster Management remain the main big issues where these technologies are highly required. Data fusion methods and tools have been adapted progressively to new data sources, which are augmenting in volume, variety, and in quick accessibility. This Special Issue gives a snapshot of the current status of that adaptation, as well as looking at what challenges are coming soon.

show abstract

“…Pohl and Van Genderen [43], Zhang [44], and Zhu et al [45] catalogued and reviewed various approaches for satellite image fusion such as intensity-hue-saturation [46], principal component analysis [47], wavelet decomposition [48], high-pass filter (HPF) [49], sparse representation [50] and area-to-point regression kriging (ATPRK) methods [51]. Recently, machine learning techniques such as deep learning [52] and random forest (RF) [53] have gained popularity in satellite image fusion. Considering its advantages and past performance [54], an RF algorithm is proposed here for the multi-sensor data alignment in the virtual constellation of L8, S2A, and ASTER.…”

Section: Introductionmentioning

confidence: 99%

Consistent Long-Term Monthly Coastal Wetland Vegetation Monitoring Using a Virtual Satellite Constellation

2021

View full text Add to dashboard Cite

Long-term monthly coastal wetland vegetation monitoring is the key to quantifying the effects of natural and anthropogenic events, such as severe storms, as well as assessing restoration efforts. Remote sensing data products such as Normalized Difference Vegetation Index (NDVI), alongside emerging data analysis techniques, have enabled broader investigations into their dynamics at monthly to decadal time scales. However, NDVI data suffer from cloud contamination making periods within the time series sparse and often unusable during meteorologically active seasons. This paper proposes a virtual constellation for NDVI consisting of the red and near-infrared bands of Landsat 8 Operational Land Imager, Sentinel-2A Multi-Spectral Instrument, and Advanced Spaceborne Thermal Emission and Reflection Radiometer. The virtual constellation uses time-space-spectrum relationships from 2014 to 2018 and a random forest to produce synthetic NDVI imagery rectified to Landsat 8 format. Over the sample coverage area near Apalachicola, Florida, USA, the synthetic NDVI showed good visual coherence with observed Landsat 8 NDVI. Comparisons between the synthetic and observed NDVI showed Root Mean Squared Error and Coefficient of Determination (R2) values of 0.0020 sr−1 and 0.88, respectively. The results suggest that the virtual constellation was able to mitigate NDVI data loss due to clouds and may have the potential to do the same for other data. The ability to participate in a virtual constellation for a useful end product such as NDVI adds value to existing satellite missions and provides economic justification for future projects.

show abstract

Fusion of SAR and Multispectral Images Using Random Forest Regression for Change Detection

Cited by 51 publications

References 38 publications

Multilayer Perceptron-Based Phenological and Radiometric Normalization for High-Resolution Satellite Imagery

Multilayer Perceptron-Based Phenological and Radiometric Normalization for High-Resolution Satellite Imagery

Multi-Source Geo-Information Fusion in Transition: A Summer 2019 Snapshot

Consistent Long-Term Monthly Coastal Wetland Vegetation Monitoring Using a Virtual Satellite Constellation

Contact Info

Product

Resources

About