In computer vision, one is often confronted with problems of domain shifts, which occur when one applies a classifier trained on a source dataset to target data sharing similar characteristics (e.g. same classes), but also different latent data structures (e.g. different acquisition conditions). In such a situation, the model will perform poorly on the new data, since the classifier is specialized to recognize visual cues specific to the source domain. In this work we explore a solution, named DeepJDOT, to tackle this problem: through a measure of discrepancy on joint deep representations/labels based on optimal transport, we not only learn new data representations aligned between the source and target domain, but also simultaneously preserve the discriminative information used by the classifier. We applied DeepJDOT to a series of visual recognition tasks, where it compares favorably against state-of-the-art deep domain adaptation methods.

International audienceDesigning an effective criterion to select a subset of features is a challenging problem for hyperspectral image classification. In this paper, we develop a feature selection method to select a subset of class discriminant features for hyperspectral image classification. First, we propose a new class separability measure based on the surrogate kernel and Hilbert Schmidt independence criterion in the reproducing kernel Hilbert space. Second, we employ the proposed class separability measure as an objective function and we model the feature selection problem as a continuous optimization problem using LASSO optimization framework. The combination of the class separability measure and the LASSO model allows selecting the subset of features that increases the class separability information and also avoids a computationally intensive subset search strategy. Experiments conducted with three hyperspectral data sets and different experimental settings show that our proposed method increases the classification accuracy and outperforms the state-of-the-art methods

International audienc

Accurate generation of a land cover map using hyperspectral data is an important application of remote sensing. Multiple classifier system (MCS) is an effective tool for hyperspectral image classification. However, most of the research in MCS addressed the problem of classifier combination, while the potential of selecting classifiers dynamically is least explored for hyperspectral image classification. The goal of this paper is to assess the potential of dynamic classifier selection/dynamic ensemble selection (DCS/DES) for classification of hyperspectral images, which consists in selecting the best (subset of) optimal classifier(s) relative to each input pixel by exploiting the local information content of the image pixel. In order to have an accurate as well as computationally fast DCS/DES, we proposed a new DCS/DES framework based on extreme learning machine (ELM) regression and a new spectral-spatial classification model, which incorporates the spatial contextual information by using the Markov random field (MRF) with the proposed DES method. The proposed classification framework can be considered as a unified model to exploit the full spectral and spatial information. Classification experiments carried out on two different airborne hyperspectral images demonstrate that the proposed method yields a significant increase in the accuracy when compared to the state-of-the-art approaches.Index Terms-Dynamic classifier selection, dynamic ensemble selection, hyperspectral image classification, markov random field model, multiple classifier system, spectral-spatial classification.

No abstract

No abstract

Land cover mapping has benefited a lot from the introduction of the Geographic Object-Based Image Analysis (GEOBIA) paradigm, that allowed to move from a pixelwise analysis to a processing of elements with richer semantic content, namely objects or regions. However, this paradigm requires to define an appropriate scale, that can be challenging in a large-area study where a wide range of landscapes can be observed. We propose here to conduct the multiscale analysis based on hierarchical representations, from which features known as differential attribute profiles are derived over each single pixel. Efficient and scalable algorithms for construction and analysis of such representations, together with an optimized usage of the random forest classifier, provide us with a semi-supervised framework in which a user can drive mapping of elements such as Small Woody Features at a very large area. Indeed, the proposed open-source methodology has been successfully used to derive a part of the High Resolution Layers (HRL) product of the Copernicus Land Monitoring service, thus showing how the GEOBIA framework can be used in a big data scenario made of more than 38,000 Very High Resolution (VHR) satellite images representing more than 120 TB of data.

Deep neural networks have established as a powerful tool for large scale supervised classification tasks. The state-of-the-art performances of deep neural networks are conditioned to the availability of large number of accurately labeled samples. In practice, collecting large scale accurately labeled datasets is a challenging and tedious task in most scenarios of remote sensing image analysis, thus cheap surrogate procedures are employed to label the dataset. Training deep neural networks on such datasets with inaccurate labels easily overfits to the noisy training labels and degrades the performance of the classification tasks drastically. To mitigate this effect, we propose an original solution with entropic optimal transportation. It allows to learn in an end-to-end fashion deep neural networks that are, to some extent, robust to inaccurately labeled samples. We empirically demonstrate on several remote sensing datasets, where both scene and pixel-based hyperspectral images are considered for classification. Our method proves to be highly tolerant to significant amounts of label noise and achieves favorable results against state-of-the-art methods.