Plenary lecture 1: concurrent neural classifiers for pattern recognition with applications in biometrics, satellite imagery, and autonomous navigation
Victor-Emil Neagoe · Annual Conference on Computers · 2009
We present the model of Concurrent Neural Classifiers (CNC) representing a collection of small neural networks, which use a global winner-takes-all strategy. Each neural module is trained to correctly classify the patterns of one class only and the number of modules equals the number M of classes. One considers the case of choosing the SOM (Self-Organized-Map) as a neural module. The CNC training technique is a supervised one, but for any individual net, the SOM specific unsupervised training algorithm is used. We built M training pattern sets and each neural module is trained with the pattern set characterized by the corresponding class label. Several presented CNC applications are dedicated to biometrics; first one has as target the recognition of color facial images and second belongs to iris recognition. One also considers a CNC application corresponding to the case of decision fusion by implementation of a multimodal biometric model. Second series of applications focuse on the CNC model for pattern recognition in multispectral satellite imagery. The implemented neural classifiers are evaluated using some LANDSAT ETM+ images composed by a set of multispectral pixels, each pixel corresponding to one of several categories (vegetation, buildings, water, and so on). Third kind of considered CNC applications correspond to visual identification of road direction of an autonomous vehicle. We present the experimental results obtained by computer simulation. We have also performed, trained and tested a real time neural path follower based on CNC model, implemented on a mobile robot (car toy).