MKL for remote sensing image classification

We propose a strategy for land use classification which exploits Multiple Kernel Learning (MKL) to automatically determine a suitable combination of a set of features without requiring any heuristic knowledge about the classification task. In this paper We present a novel procedure that allows MKL to achieve good performance in the case of small training sets. Experimental results on publicly available datasets demonstrate the feasibility of the proposed approach.

Scheme of the four main steps that form the proposed heuristic for the selection of the kernels

Weights selected by the heuristic MKL when trained on the 21-class data set. Numbers at the bottom indicate kernels: 1 for linear, 2–5 for RBFs , and 6–9 for chi-square

Weights selected by the heuristic MKL when trained on the 19-class data set. Numbers at the bottom indicate kernels: 1 for linear, 2–5 for RBFs , and 6–9 for chi-square

Features

Each zip file (.zip) contains the features (in matlab format .mat) used for the classification of the 21-class and 19-class datasets: Bag of SIFT, Gist, Bag of LBP, LBP of dense moments.

Dataset Split

Each zip file contains the splits of both training and test data of the 21-class and 19-class datasets. Each matlab file (*_idx_100.mat) contains 100 splits at different percentages of the training and test set: 5, 10, 20, 50, 80, 90. The matlab file (labels.mat) contains the labels of each image.

Publications

Remote Sensing Image Classification Exploiting Multiple Kernel Learning
(Claudio Cusano, Paolo Napoletano, Raimondo Schettini) In IEEE Geoscience and Remote Sensing Letters, volume 12, number 11, pp. 2331-2335, 2015.

Download BibTex Doi

@article{cusanoremote,
 author = {Cusano, Claudio and Napoletano, Paolo and Schettini, Raimondo},
 year = {2015},
 month = {1},
 pages = {2331-2335},
 title = {Remote Sensing Image Classification Exploiting Multiple Kernel Learning},
 volume = {12},
 number = {11},
 journal = {IEEE Geoscience and Remote Sensing Letters},
 keywords = {geophysical, techniques, image, classification, land, use, remote, sensing, kernel, learning, land, use, classification, remote, sensing, image, classification, Accuracy, Kernel, Optimization, Remote, sensing, Satellites, Standards, Training, Multiple, kernel, learning, MKL, remote, sensing, image, classification},
 pdf = {/download/cusano2015remote-sensing.pdf},
 doi = {10.1109/LGRS.2015.2476365},
 issn = {1545-598X}}