Target Recognition Method Research Based On Ultra-Wide Band Signals Under Foliage Enviroments

Target recognition under foliage environment is a research topic of great significance for both civil and military applications. However, due to the special nature of the environment foliage that, like great number of trees and lush foliage, it would cause serious sight blocked. What’s more, the ambient noise and multipath fading effect will also bring a lot of interference to signal communication. These make the traditional sophisticated pattern recognition techniques such as image identification, infrared recognition not meet the requirements under this environment. The ultra-wideband (UWB) technology with high resolution in the time domain, and a good ability to penetrate obstacles has broad application prospects in many fields like obstacle detection, imaging through walls, underground target detection. It is very suitable for target recognition research under foliage environments.In this paper, making use of ultra-wide band’s immunity to high noise and multipath effects, and strong penetration ability to accomplish the target identification purpose while communicating through ultra-wideband signals. This study is aimed to identify the number of targets, namely the number of human as detectors in the target environment. There are some researches related to using ultra-wideband technology under foliage environmental on recognizing people, wood and metal box. Taking into account the 3 kinds of targets having different dielectric constant due to different material, which leads to very different impact on ultra-wideband channel, it has pretty good classification results. But when the targets belong to the same kind, it would be a problem whether it recognize the targets’number effectively.By building test scenarios in real foliage environment collecting data to validate the target recognition test. Each scene has a different number of targets respectively, stationing in randomly positions to test several times achieving the random requirement. After obtaining a sufficient number of sample data, ultra-wideband signal received wavelet transform to extract the wavelet coefficients of the received signal, and then use kernel principal component analysis on these wavelet coefficients to obtain dimension reduction and remove redundant information. These principle components are used as features to characterize the channel characteristics, and combine them with PSO-ACO optimized support vector machine combined as the SVM input vector, while the output is the classification result. Using the training sample set to train the SVM model, and taking advantage of the support vector machine’s learning and classification ability to classify the test samples to achieve the goal that discriminate the target number well.