The acronym IPARP stands for “Integrated Pattern Analysis, Recognition, and Prediction.” The toolbox was created in Matlab (version 5.2) to facilitate performance analysis common in data mining and automatic target recognition (ATR) applications. The toolbox provides an integrated, semi-automated environment in which ATR researchers can address various “what-if” questions in a straightforward and systematic manner. Some typical queries that can be answered by the toolbox are as follows:
q How many features do I need for maximum ATR performance?
q What classifier should I use for this particular problem?
q Where is the point of diminishing returns?
q What is the impact of data mismatch on data-mining performance?
q Do data statistics change over time?
q What is the performance-versus-cost trade-off if I substitute computationally expensive features with something less expensive using approximation algorithms?
q Are there features that work consistently well across multiple environmental conditions?
q Is in situ performance optimization necessary? If yes, what is the appropriate classifier/feature structure?
The toolbox is designed to reduce the algorithm-design time so that we can focus on more important and interesting aspects of algorithm design—finding innovative ways to solve challenging problems by quickly knowing what works, what doesn’t, and why. In short, the goal of designing this toolbox was to enable the user to rapidly optimize the performance of a complex ATR system.
ps: 最近写的一个报告中，涉及到了自动目标识别方面的知识，查了很多文献，感觉IPARP Toolbox能从整体上反映了我们的思想，有些不谋而合的感觉。For more details, http://www.killmine.com/OnlineManual/TechnologyOverview.htm
posted on 2007-10-17 22:02
N度自由风的c++小窝 阅读(288)
评论(0) 编辑 收藏 引用