Less is more: optimizing classification performance through feature selection in a very-high-resolution remote sensing object-based urban application

Georganos, Stefanos; Grippa, Taïs; Vanhuysse, Sabine; Lennert, Moritz; Shimoni, Michal; Kalogirou, Stamatis; Wolff, Éléonore

doi:10.1080/15481603.2017.1408892

Cited by 191 publications

(136 citation statements)

References 46 publications

Supporting

Mentioning

130

Contrasting

Unclassified

Order By: Relevance

“…While these images typically feature few spectral bands (normally optical RGB plus potentially one infrared band), limiting their ability to differentiate land cover types based on spectral characteristics, their high resolution enables the human eye to differentiate most land cover types based on, for example, the shape and density of trees, and the advancement of object-based classification methods has meant that automated processes can also take advantage of this spatial information to produce accurate classifications [13][14][15][16]. However, the high costs and complexity of both the object-based image analysis, and the high cost and low availability of data at a sufficient resolution, remain as challenges for wider application [17,18]. Our study investigated whether publicly available data, namely Sentinel-2 (S2), can map complex landscapes in Southern Myanmar, including oil palm, rubber, and betel nut plantations using a Random Forest classifier on Google Earth Engine.…”

Section: Introductionmentioning

confidence: 99%

More Than Meets the Eye: Using Sentinel-2 to Map Small Plantations in Complex Forest Landscapes

Nomura

Mitchard

2018

Remote Sensing

View full text Add to dashboard Cite

Many tropical forest landscapes are now complex mosaics of intact forests, recovering forests, tree crops, agroforestry, pasture, and crops. The small patch size of each land cover type contributes to making them difficult to separate using satellite remote sensing data. We used Sentinel-2 data to conduct supervised classifications covering seven classes, including oil palm, rubber, and betel nut plantations in Southern Myanmar, based on an extensive training dataset derived from expert interpretation of WorldView-3 and UAV data. We used a Random Forest classifier with all 13 Sentinel-2 bands, as well as vegetation and texture indices, over an area of 13,330 ha. The median overall accuracy of 1000 iterations was >95% (95.5%-96.0%) against independent test data, even though the tree crop classes appear visually very similar at a 20 m resolution. We conclude that the Sentinel-2 data, which are freely available with very frequent (five day) revisits, are able to differentiate these similar tree crop types. We suspect that this is due to the large number of spectral bands in Sentinel-2 data, indicating great potential for the wider application of Sentinel-2 data for the classification of small land parcels without needing to resort to object-based classification of higher resolution data.

show abstract

Section: Introductionmentioning

confidence: 99%

More Than Meets the Eye: Using Sentinel-2 to Map Small Plantations in Complex Forest Landscapes

Nomura

Mitchard

2018

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…SSPO methods have been criticized for being cost-ineffective since they operate on a trial-and-error manner (for visual interpretation) or require reference segments to be digitized (for quantitative analysis), and also for being susceptible to the subjectivity of the user. To treat this issue, unsupervised segmentation parameter optimization (USPO) methods have been developed and are particularly important in the context of increasing data loads and automation purposes [14,[18][19][20][21][22][23][24][25]. To identify optimal segmentation parameters, USPO procedures usually employ a combination of geospatial metrics that describe spectral heterogeneity between and within image segments [9,[26][27][28].…”

Section: Introductionmentioning

confidence: 99%

Normalization in Unsupervised Segmentation Parameter Optimization: A Solution Based on Local Regression Trend Analysis

et al. 2018

Self Cite

View full text Add to dashboard Cite

In object-based image analysis (OBIA), the appropriate parametrization of segmentation algorithms is crucial for obtaining satisfactory image classification results. One of the ways this can be done is by unsupervised segmentation parameter optimization (USPO). A popular USPO method does this through the optimization of a "global score" (GS), which minimizes intrasegment heterogeneity and maximizes intersegment heterogeneity. However, the calculated GS values are sensitive to the minimum and maximum ranges of the candidate segmentations. Previous research proposed the use of fixed minimum/maximum threshold values for the intrasegment/intersegment heterogeneity measures to deal with the sensitivity of user-defined ranges, but the performance of this approach has not been investigated in detail. In the context of a remote sensing very-high-resolution urban application, we show the limitations of the fixed threshold approach, both in a theoretical and applied manner, and instead propose a novel solution to identify the range of candidate segmentations using local regression trend analysis. We found that the proposed approach showed significant improvements over the use of fixed minimum/maximum values, is less subjective than user-defined threshold values and, thus, can be of merit for a fully automated procedure and big data applications.

show abstract

“…Since they are sufficient for performing the feature selection step, this would save processing time and storage space [61]. Only the most discriminant features could then be computed for the whole AOI.…”

Section: Discussionmentioning

confidence: 99%

Mapping Urban Land Use at Street Block Level Using OpenStreetMap, Remote Sensing Data, and Spatial Metrics

Grippa

Georganos

Zarougui

et al. 2018

IJGI

Self Cite

View full text Add to dashboard Cite

Up-to-date and reliable land-use information is essential for a variety of applications such as planning or monitoring of the urban environment. This research presents a workflow for mapping urban land use at the street block level, with a focus on residential use, using very-high resolution satellite imagery and derived land-cover maps as input. We develop a processing chain for the automated creation of street block polygons from OpenStreetMap and ancillary data. Spatial metrics and other street block features are computed, followed by feature selection that reduces the initial datasets by more than 80%, providing a parsimonious, discriminative, and redundancy-free set of features. A random forest (RF) classifier is used for the classification of street blocks, which results in accuracies of 84% and 79% for five and six land-use classes, respectively. We exploit the probabilistic output of RF to identify and relabel blocks that have a high degree of uncertainty. Finally, the thematic precision of the residential blocks is refined according to the proportion of the built-up area. The output data and processing chains are made freely available. The proposed framework is able to process large datasets, given that the cities in the case studies, Dakar and Ouagadougou, cover more than 1000 km 2 in total, with a spatial resolution of 0.5 m.

show abstract

Less is more: optimizing classification performance through feature selection in a very-high-resolution remote sensing object-based urban application

Cited by 191 publications

References 46 publications

More Than Meets the Eye: Using Sentinel-2 to Map Small Plantations in Complex Forest Landscapes

More Than Meets the Eye: Using Sentinel-2 to Map Small Plantations in Complex Forest Landscapes

Normalization in Unsupervised Segmentation Parameter Optimization: A Solution Based on Local Regression Trend Analysis

Mapping Urban Land Use at Street Block Level Using OpenStreetMap, Remote Sensing Data, and Spatial Metrics

Contact Info

Product

Resources

About