Arnaud Le Bris

Use intermediate results of wrapper band selection methods: A first step toward the optimization of spectral configuration for land cover classifications

Intermediate results of two state-of-the-art wrapper feature selection approaches (GA and SFFS) associated to a classifier (linear SVM) applied to hyperspectral data sets were used to derive information about band importance for specific land cover classification problems. The impact of the number of selected bands on classification accuracy was obtained thanks to SFFS, while a band importance measure was derived from intermediate sets of bands tested by GA. Such results are a first step toward the identification of the most suitable spectral bands to design superspectral camera systems dedicated to specific applications (e.g. classification of urban land cover and material maps).

Comparison of VHR panchromatic texture features for tillage mapping

Agricultural practices are major drivers of water flows in cultivated landscapes. Especially, the spatial arrangements and connectivities of tilled/untilled fields have a strong impact onto run off and soil erosion at the landscape and watershed scales. Very high spatial resolution satellite images offer the possibility to classify tilled vs. untilled fields at a large scale. This paper compares the importance of various VHR texture features for tillage mapping. Classical texture features such as coocurrence Haralick descriptors, Gabor and SIFT-based descriptors are studied. The random forest classifier is used to assess feature importance. A 50 cm panchromatic WorldView-I image is used for experiments. Very good classification accuracies of 83.4 % and 94.5 % are reached.

Contribution of band selection and fusion for hyperspectral classification

For some specific land cover classification problems, it may be interesting to design superspectral camera systems with reduced numbers of bands (∼ 20) and optimized band widths. This paper assesses the contribution of band selection and band fusion processes separately and jointly for dimensionality reduction. The proposed approach is fully automatic and based on a wrapper feature selection using Random forest classifier and a similarity-based fusion process. While combining both processes, selection before fusion gave the best results, reducing by almost 91% the number of bands while keeping satisfying accuracies. Results are presented on Indian Pines, Salinas and Pavia Centre hyperspectral datasets.

A random forest class memberships based wrapper band selection criterion: Application to hyperspectral

Hyperspectral imagery generates huge data volumes, consisting of hundreds of contiguous and often highly redundant spectral bands. Difficulties are caused by this high dimensionality. Feature selection (FS) is a possible strategy to reduce the number of bands, consisting in selecting the most relevant bands for a classification problem. It is adapted to the design of superspectral sensor dedicated to specific applications. FS is an optimization problem involving both a metric (that is to say a FS score or criterion measuring the relevance of feature subsets) to optimize and an optimization strategy. In this paper, a wrapper FS score based on Random Forests (RF) and taking into account RF class membership measures was proposed. It was compared to a state-of-the-art wrapper FS score (classification Kappa obtained by RF). Both were then evaluated quantitatively considering both classification performance reached applying different classifiers. An qualitative analysis was also performed to consider the stability/regularity of the selected features along the spectrum. Even though the quantitative evaluation showed little differences between the two tested FS criteria, there seemed to be a trend in favour of the proposed criterion. Taking into account the measures of class membership provided by a RF classifier slightly improved results, regularizing feature selection.

Combining top-down and bottom-up approaches for building detection in a single very high resolution satellite image

Building detection from geospatial optical images has been a popular topic of research for the last twenty years and in particular with the emergence of very high resolution satellites. Existing methods exhibit various flaws and prevent them from being efficient at large scales of space and time: they are context-dependent, require a tedious parameter tuning or several data sources. In this paper, we propose a fully automatic method that alleviates some of these issues by combining the strengths of bottom-up and top-down approaches, i.e., of both classification and pattern recognition algorithms. This allows to correctly detect the objects by geometric prior knowledge while finely delineating their borders and preserving their shapes. The method is evaluated over a complex area of more than 230 buildings using a 0.5 m multispectral pansharpened Pleiades image.

Identify important spectrum bands for classification using importances of wrapper selection applied to hyperspectral data

Intermediate results of two state-of-the-art wrapper feature selection approaches (GA and SFFS) applied to hyperspectral data sets were used to derive information about band importance for specific land cover classification problems. Several feature selection performance scores (classification accuracies, Bhattacharyya separability) were tested. The impact of the number of selected bands on classification accuracy was obtained thanks to SFFS, while a band importance measure was derived from intermediate sets of bands tested by GA. Such results are a first step toward the identification of the most suitable spectral bands to design superspectral camera systems dedicated to specific applications (e.g. classification of urban land cover and material maps).

Optical Remote Sensing in Urban Environments

Abstract: Cities today face a variety of issues: attractiveness and economic development, living conditions and urban redevelopment, the quality of life of citizens and the environmental conditions of the urban system as a whole.

Agricultural field delimitation using active learning and random forests margin

Agricultural practices and spatial arrangements of fields have a strong impact on water flows in cultivated landscapes. In order to monitor landscapes at a large scale, there is a strong need for automatic or semi-automatic field delineation. Field measurements for delineating parcel network are not efficient, thus very high resolution satellite imagery should help delineating agricultural fields in a automatic way. This study focuses on agricultural field delineation based on the classification of very high resolution satellite imagery. A hybrid approach is proposed and combines a region-based approach and active learning (AL) techniques. Random forest (RF) classifier is used for classification and feature selection. The margin concept is used as uncertainty measure in active learning algorithm. Satisfying results are shown on a Geoeye image. AL RF model is compared to simple and global RF models that are built from adjacent and geographically distant fields respectively.

Hierarchically exploring the width of spectral bands for urban material classification

In urban areas, material maps, i.e. knowledge concerning the roofing materials or the different kinds of ground areas, are necessary for several city modeling or monitoring applications. Airborne remote sensing techniques appear to be convenient for providing them at a large scale but require an enhanced imagery spectral resolution. A superspectral sensor with a limited number of bands dedicated to urban materials classification could be a solution. Within this context, this study focused on the optimization of this band subset from hyperspectral data, considering both the position of the bands and their width. The used approach first builds a hierarchy of groups of adjacent bands, according to a relevance criterion to decide which adjacent bands must be merged. Then, band selection is performed at the different levels of this hierarchy. Several band configurations are thus explored within this hierarchy. This method was applied to a data set consisting of spectra generated from reflectance spectral signatures of 9 common urban materials collected from 7 spectral libraries. At the end, the potential of a superspectral sensor with wider bands was confirmed.

Fully automatic analysis of archival aerial images current status and challenges

Archival aerial images are a unique and relatively unexplored means to generate detailed land-cover information in 3D over the past 100 years. Many long-term environmental monitoring studies can be based on this type of image series. Such data provide a relatively dense temporal sampling of the territories with very high spatial resolution. Furthermore, photogrammetric workflows exist in order to both produce orthoimages and Digital Surface Models, with reasonable interactive actions. However, today, there is no fully automatic pipeline for generating such kind of data. This paper presents the main avenues of research in order to develop such workflow, starting from registration and radiometric issues up to land-cover classification challenges.