Building a Cascade Detector and Applications in Automatic Target Detection

A hierarchical classifier (cascade) is proposed for target detections. In building an optimal cascade, three heuristics are considered: (1) frontier following approximation, (2) controlling error rates, and (3) weighting. Simulations on synthetic data with various underlying distributions are carried out. It is found that weighting heuristic is optimal in both computational complexity and error rates. The contribution of this paper is to initiate a systematic comparison of several potential heuristics that can be utilized in building a hierarchical model. A range of discussions – regarding the implications and the promises of the cascade architecture, as well as techniques that can be integrated into this framework – will be provided. The current winner – weighting algorithms – is applied to an IR data set. It is found that it outperforms some state-of-the-art approaches that utilize the same type of simple classifiers.