Classification of hyperspectral imagery with neural networks: comparison to conventional tools
- PDF / 2,505,712 Bytes
- 19 Pages / 595 x 794 pts Page_size
- 59 Downloads / 212 Views
R ESEA R CH
Open Access
Classification of hyperspectral imagery with neural networks: comparison to conventional tools Erzsébet Merényi1,2* , William H Farrand3 , James V Taranik4 ˆ and Timothy B Minor4
Abstract Efficient exploitation of hyperspectral imagery is of great importance in remote sensing. Artificial intelligence approaches have been receiving favorable reviews for classification of hyperspectral data because the complexity of such data challenges the limitations of many conventional methods. Artificial neural networks (ANNs) were shown to outperform traditional classifiers in many situations. However, studies that use the full spectral dimensionality of hyperspectral images to classify a large number of surface covers are scarce if non-existent. We advocate the need for methods that can handle the full dimensionality and a large number of classes to retain the discovery potential and the ability to discriminate classes with subtle spectral differences. We demonstrate that such a method exists in the family of ANNs. We compare the maximum likelihood, Mahalonobis distance, minimum distance, spectral angle mapper, and a hybrid ANN classifier for real hyperspectral AVIRIS data, using the full spectral resolution to map 23 cover types and using a small training set. Rigorous evaluation of the classification accuracies shows that the ANN outperforms the other methods and achieves ≈90% accuracy on test data. Keywords: Classification; Hyperspectral imagery; Neural networks; High-dimensional data
1 Introduction High spatial and spectral resolution images from advanced remote sensors such as NASA’s AVIRIS (e.g., [1]), Hyperion, HyMap, HYDICE [2], and others provide abundant information for the understanding and monitoring of the Earth. At the same time, they produce data of unprecedented volume and complexity. Unraveling important processes such as the evolution of the solid earth, global cycling of energy, oxygen, water, etc., the responses of the biosphere to disturbances, and others mandates the best possible exploitation of the data. The challenge is to develop methods that are powerful enough to make use of the intricate details in hyperspectral data and are fast, robust, noise tolerant, and adaptive. While the growing number of spectral channels enables *Correspondence: [email protected] ˆDeceased 1 Department of Statistics, Rice University, 6100 Main Street MS-138, Houston, TX 77005, USA 2 Department of Electrical and Computer Engineering, Rice University, 6100 Main Street, Houston, TX 77005, USA Full list of author information is available at the end of the article
discrimination among a large number of cover classes, many conventional techniques fail on these data because of mathematical or practical limitations. For example, the maximum likelihood and other covariance-based classifiers require, on the minimum, as many training samples per class as the number of bands plus one, which creates a severe problem of field sampling for AVIRIS 224-channel data with many classes. Dimensionality reduction is
Data Loading...
