Lionel Gueguen

Classifying Compound Structures in Satellite Images: A Compressed Representation for Fast Queries

With the increased spatial resolution of current sensor constellations, more details are captured about our changing planet, enabling the recognition of a greater range of land use/land cover classes. While pixeland object-based classification approaches are widely used for extracting information from imagery, recent studies have shown the importance of spatial contexts for discriminating more specific and challenging classes. This paper proposes a new compact representation for the fast query/classification of compound structures from very high resolution optical remote sensing imagery. This bag-of-features representation relies on the multiscale segmentation of the input image and the quantization of image structures pooled into visual word distributions for the characterization of compound structures. A compressed form of the visual word distributions is described, allowing adaptive and fast queries/classification of image patterns. The proposed representation and the query methodology are evaluated for the classification of the UC Merced 21-class data set, for the detection of informal settlements and for the discrimination of challenging agricultural classes. The results show that the proposed representation competes with state-of-the-art techniques. In addition, the complexity analysis demonstrates that the representation requires about 5% of the image storage space while allowing us to perform queries at a speed down to 1 s/ 1000 km2/CPU for 2-m multispectral data.

Large-scale damage detection using satellite imagery

Satellite imagery is a valuable source of information for assessing damages in distressed areas undergoing a calamity, such as an earthquake or an armed conflict. However, the sheer amount of data required to be inspected for this assessment makes it impractical to do it manually. To address this problem, we present a semi-supervised learning framework for large-scale damage detection in satellite imagery. We present a comparative evaluation of our framework using over 88 million images collected from 4, 665 KM2 from 12 different locations around the world. To enable accurate and efficient damage detection, we introduce a novel use of hierarchical shape features in the bags-of-visual words setting. We analyze how practical factors such as sun, sensor-resolution, satellite-angle, and registration differences impact the effectiveness our proposed representation, and compare it to five alternative features in multiple learning settings. Finally, we demonstrate through a user-study that our semi-supervised framework results in a ten-fold reduction in human annotation time at a minimal loss in detection accuracy compared to manual inspection.

Toward a Generalizable Image Representation for Large-Scale Change Detection: Application to Generic Damage Analysis

Each year, multiple catastrophic events impact vulnerable populations around the planet. Assessing the damage caused by these events in a timely and accurate manner is crucial for efficient execution of relief efforts to help the victims of these calamities. Given the low accessibility of the damaged areas, high-resolution optical satellite imagery has emerged as a valuable source of information to quickly asses the extent of damage by manually analyzing the pre- and postevent imagery of the region. To make this analysis more efficient, multiple learning techniques using a variety of image representations have been proposed. However, most of these representations are prone to variabilities in capture angle, sun location, and seasonal variations. To evaluate these representations in the context of damage detection, we present a benchmark of 86 pre- and postevent image pairs with respective reference data derived from United Nation Operational Satellite Applications Programme (UNOSAT) assessment maps, spanning a total area of 4665 km2 from 11 different locations around the world. The technical contribution of our work is a novel image representation based on shape distributions of image patches encoded with locality-constrained linear coding. We empirically demonstrate that our proposed representation provides an improvement of at least 5%, in equal error rate, over alternate approaches. Finally, we present a thorough robustness analysis of the considered representational schemes, with respect to capture-angle variabilities and multiple sensor combinations.

Image patch characterization with shape distributions: Application to WorldView-2 images

This paper describes an image patch characterization for image information mining tasks. An image patch is first decomposed into a multi-scale segmentation thanks to the Max Tree representation. Then, each segment is described by shift invariant shape attributes. Finally, the segment attributes are aggregated into a shape distribution which constitutes the patch characterization. Illustrations of this image content description are given for patches of a WorldView-2 multi-spectral scene, and the information relevance is assessed by an automatic classification of the patch characteristics which is compared to land use/land cover annotations.

Mapping people distribution at very high resolution from remote sensing imagery and a priori coarse people counts

The paper presents a fast and fully automatic method for estimating people distribution maps at very high resolution (≥ 25 m) from VHR optical imagery. This method is implemented in the High-res Urban Globe (HUG) suite of tools developed at DigitalGlobe, Inc. The methods relies on the a priori knowledge of coarse people counts to train a model acting on building features extracted from VHR imagery. Experiments and results show the high fidelity of HUG population density estimate to detailed census data.

Interscale learning and classification for global HR/VHR image information extraction

An interscale learning paradigm for global HR/VHR image information extraction is presented. The paradigm relies on the information matching between a priori global knowledge and features derived from high resolution imagery to perform adaptive high resolution land classification. Unlike traditional machine learning techniques, this strategy avoids the costly collection of local training datasets and the local parameter tuning and it enables the full automation of the information extraction process.

The Local Mutual Information invariance: Application to change detection between multispectral images

This paper proposes an experimental analysis of the Local Mutual Information (LMI) invariance in the context of unsupervised change detection between multi-spectral images. The conversion from Digital Numbers to Surface Reflectance does not affect theoretically the LMI based change map under the assumption of homogeneous atmospheric conditions. Experiments are conducted with QuickBird bi-temporal images to assess the robustness of the LMI to conversion to Surface Reflectance. Results show that the probability estimation, upon which the LMI depends on, is critical for maintaining invariant LMI based change detection values.

Very High Spatial Resolution Optical Imagery: Tree-Based Methods and Multi-temporal Models for Mining and Analysis

This chapter presents a selection of image representation methods for very high spatial resolution optical data acquired by agile satellite platforms. Each method aims at specific properties of the image information content and can be tailored to address unique features in spatial, temporal, and angular acquisitions. Techniques for the identification and characterization of surface structures and objects often employ spatial and spectral features best represented in panchromatic and multi-spectral images, respectively. In both cases, the vastness of the data space can only be addressed effectively by means of some data representation structure that organizes the image information content in meaningful ways. The latter suggest that a globally optimal representation of the object(s) of interest can be obtained through interactions with a scale space as opposed to single-scale information layer. Two examples, the Max-Tree and Alpha-Tree algorithms, are discussed in the context of interactive big data information mining. Optical and structural properties of the surface materials can be exploited by analyzing the tempo-angular domain by means of anisotropic decompositions, which rely on the availability of surface reflectance data and dense angular sampling. Bidirectional reflectance distribution functions of various materials are discussed in detail, showing that the temporal information should always be coupled to the corresponding angular component to make the best use of the available imagery.

A comparison of land use land cover classification using superspectral WorldView-3 vs hyperspectral imagery

In advance of releasing a WorldView-3 (WV-3) dataset with both VNIR and SWIR bands for research purposes, this study was conducted to provide a baseline comparison of land use/land cover (LULC) classification based on hyperspectral and 16-, 8-, and 4-bands of WV-3 imagery. We chose a well-researched area over the city center of Pavia, Italy. Results suggest that the addition of spectral information from WV-3s SWIR bands helps bridge the gap between precision/recall scores obtained with multispectral VNIR vs. hyperspectral VNIR imagery.