|
|
|
“Fuzzy Logic for Unattended Ground Sensor Fusion”
by Arjun Bharadwaj and Jerry M. Mendel
January 2006
This report summarizes our research from Jan ’05 to Dec ’05. We have developed multi-category classifiers based on seismic data to classify heavy-tracked, heavy wheeled, light tracked and light-wheeled vehicles. We focused on data collected in the normal terrain.
We also developed fusion algorithms for type-1 and type-2 Fuzzy Logic Rule Based Classifiers (FLRBCs) based on the Choquet Fuzzy Integral (CFI). We conducted experiments to evaluate the performance of the classifiers and to evaluate the effectiveness of seismic data for classification. We also conducted experiments to evaluate the performances of the fused classifiers (both acoustic and seismic) and determine if pweformance could be improved. Our results show that binary classification between tracked and wheeled vehicles is effective using seismic data. However, due to the inherent unreliability of the seismic data, the performance of the classifiers based on seismic data was poor when compared to the performance of the classifiers based on acoustic data. Fusing the two classifiers also did not show any appreciable improvement in performance.
We note that FL-RBCs performed better than the Bayesian equivalent for all the experiments. This shows that FL_RBCs are better suited to handle uncertainties in the data.