Mark Schmalz (University of Florida), Gerhard X. Ritter (University of Florida)
Keywords: NROC, Non-Resolved Object Characterization
Abstract:
Accurate multispectral or hyperspectral signature classification is key to the nonimaging detection and recognition of space objects. Additionally, signature classification accuracy depends on accurate spectral endmember determination [1]. Previous approaches to endmember computation and signature classification were based on linear operators or neural networks (NNs) expressed in terms of the algebra (R, +, x) [1,2]. Unfortunately, class separation in these methods tends to be suboptimal, and the number of signatures that can be accurately classified often depends linearly on the number of NN inputs. This can lead to poor endmember distinction, as well as potentially significant classification errors in the presence of noise or densely interleaved signatures.
In contrast to traditional CNNs, autoassociative morphological memories (AMM) are a construct similar to Hopfield autoassociatived memories defined on the (R, +, ?,?) lattice algebra [3]. Unlimited storage and perfect recall of noiseless real valued patterns has been proven for AMMs [4]. However, AMMs suffer from sensitivity to specific noise models, that can be characterized as erosive and dilative noise. On the other hand, the prior definition of a set of endmembers corresponds to material spectra lying on vertices of the minimum convex region covering the image data. These vertices can be characterized as morphologically independent patterns. It has further been shown that AMMs can be based on dendritic computation [3,6]. These techniques yield improved accuracy and class segmentation/separation ability in the presence of highly interleaved signature data.
In this paper, we present a procedure for endmember determination based on AMM noise sensitivity, which employs morphological dendritic computation. We show that detected endmembers can be exploited by AMM based classification techniques, to achieve accurate signature classification in the presence of noise, closely spaced or interleaved signatures, and simulated camera optical distortions. In particular, we examine two critical cases: (1) classification of multiple closely spaced signatures that are difficult to separate using distance measures, and (2) classification of materials in simulated hyperspectral images of spaceborne satellites. In each case, test data are derived from a NASA database of space material signatures. Additional analysis pertains to computational complexity and noise sensitivity, which are superior to classical NN based techniques.
%Z Winter, M.E., Fast autonomous spectral end-member determination in hyperspectral data, in Proceedings of the 13th International Conference On Applied Geologic Remote Sensing, Vancouver, B.C., Canada, 337-44 1999
Keshava, N., A survey of spectral unmixing algorithms, Lincoln Laboratory Journal 14:55-78, 2003
Ritter, G.X., L. Iancu, and G. Urcid, Morphological perceptrons with dendritic structure, in Proceedings of the 2003 IEEE Conference on Fuzzy Systems, pp. 1296-1301, 2003
Grana, M. , Sussner, P., Ritter, G.X., Associative morphological memories for endmember determination in spectral unmixing, Proceedings of the 12th IEEE International Conference on Fuzzy Systems, 2: 1285-1290, 2003
Ritter, G.X., Iancu, L., Schmalz, M.S., A new auto-associative memory based on lattice algebra, in Proceedings of the Ninth Iberoamerican Congress on Pattern Recognition, Puebla, Mexico, pp. 148-55, 2004
Ritter, G.X., Schmalz, M.S., Learning in lattice neural networks that employ dendritic computing, to appear in Proceedings of the 2006 IEEE World Congress on Computational Intelligence (in publication)
Date of Conference: September 10-14, 2006
Track: Non-Resolved Object Characterization