Scaling a neuro fuzzy system and applications to 3D visualization and robot path planning

We present a new scaling technique for a neurofuzzy system by implementing an equivalently reduced order system from the given data. This technique is developed for designing a reduced rule-based neurofuzzy system by simultaneously reducing the number of observations and the number of variables from the original data. Performance and accuracy of the approach have been established through the data from search time of target detection with correlation coefficients between the actual output from the lab experiment and the estimated outputs from neurofuzzy systems using reduced data sets and the original data set. First, factor analysis is used to extract a set of hidden feature vectors that are used to linearly combine the elements of the full feature vector of the original variables (measurements). Then, fuzzy c-means (FCM) clustering analysis is applied to map the clustered observations from 44 digitized color images along with the associated target and background metrics for the targets. The estimated output by the proposed algorithm is compared with the actual output from the experiment through the behaviors of neurofuzzy systems from the original data set and the reduced order data sets. The paper also discusses some possible applications of the proposed technique for robot path planning and its 3D visualization.