Study Of Target Detection And Recognition Based On UWB-IR And Neural Network

UWB-IR is a short range wireless communication method using short pulses to transmit information; it has technology advantages of strong distinguish, strong detection, and anti-jamming abilities. It has obtained significant achievements in transportation, medical and security aspects with the combination of neural networks in the field of target recognition. However, in some applications which need communication and target recognition such as individual combat, intrusion detection, UWBR technologies can only complete target detection, but not complete communication. Therefore, how to achieve target recognition with the foundation of UWB communication and neural network is the subject of this study.For the subject of above, after analyzing a large number of simulations and actual measurements, we find that when communication channel has different types of targets, the received signals will demonstrate different characters, and we may extract characteristic parameters related to targets using these differences of characters, and create the mapping model between received signals and the target with the help of neural network, so as to achieve the dual purpose of target detection and recognition and communication. Based on the above ideas, the paper focuses on the communication perspective, collects the received signals, and extracts the characteristic parameters of signals, then uses neural network to identify the type of targets in the communication channel. In order to verify the efficiency of the method, firstly, the paper establishes the theoretical simulation model using time-domain FDTD theory, and applies the method to the indoor signals propagation and target recognition, and discusses the influence of the different signal-to-noise ratio to the target recognition. The results show that the method is effective for target recognition, and the noises have different influence to the recognition results of different types of targets. Secondly, the paper collects the communication received signals in an actual small foliage measurement activity using P400, then removes noises and extracts the parameters, and the neural network is optimized by PSO to solve the randomness of the initial parameters to bring the performance degradation and further improve the recognition rate and stability. The final simulation results show that the method is still effective in the actual environment, and the method optimized by PSO has higher identification rate.