原文传递
Beyond Scheme F
| 题名： | Beyond Scheme F |
| 作者： | Gillmann, R.;Pepin, J.;Fisher, H.;Elliott, C.J.; |
| 关键词： | 32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATION ;99 MATHEMATICS, COMPUTERS, INFORMATION SCIENCE, MANAGEMENT, LAW, MISCELLANEOUS; TRAFFIC CONTROL; VEHICLES; SORTING; DATA BASE MANAGEMENT; ALGORITHMS; NEURAL NETWORKS; CLASSIFICATION; DATA ACQUISITION; MATRICES; PROBABILITY |
| 摘要： | Traffic classification techniques were evaluated using data from a 1993 investigation of the traffic flow patterns on I-20 in Georgia. First we improved the data by sifting through the data base, checking against the original video for questionable events and removing and/or repairing questionable events. We used this data base to critique the performance quantitatively of a classification method known as Scheme F. As a context for improving the approach, we show in this paper that scheme F can be represented as a McCullogh-Pitts neural network, oar as an equivalent decomposition of the plane. We found that Scheme F, among other things, severely misrepresents the number of vehicles in Class 3 by labeling them as Class 2. After discussing the basic classification problem in terms of what is measured, and what is the desired prediction goal, we set forth desirable characteristics of the classification scheme and describe a recurrent neural network system that partitions the high dimensional space up into bins for each axle separation. the collection of bin numbers, one for each of the axle separations, specifies a region in the axle space called a hyper-bin. All the vehicles counted that have the same set of in numbers are in the same hyper-bin. The probability of the occurrence of a particular class in that hyper- bin is the relative frequency with which that class occurs in that set of bin numbers. This type of algorithm produces classification results that are much more balanced and uniform with respect to Classes 2 and 3 and Class 10. In particular, the cancellation of errors of classification that occurs is for many applications the ideal classification scenario. The neural network results are presented in the form of a primary classification network and a reclassification network, the performance matrices for which are presented. |
| 总页数： | 15p |
| 报告类型： | 科技报告 |
相关文献
- ENERGY AND TRANSPORTATION BEYOND 2000.
- NASA Supply Chain Beyond Low Earth Orbit: Explore Moon to Mars.
- Optimal flight trajectories in the presence of windshear, 1984-86
- Robust Sliding Window Synchronizer Developed
- Future enhancements to ground-based microburst detection
- Probabilistic Vehicle Identification Techniques for Semiautomated Transportation Security.
检索历史
应用推荐
