Research into target recognition techniques based on device-free sensing approach

Publication Type:
Thesis
Issue Date:
2019
Full metadata record
Using wireless communication signals, also known as radio-frequency (RF) signals, for not only data communication but also target detection and classification, is an emerging device-free sensing (DFS) technology that has a great potential in our daily life. In particular, this technology can be used for target detection in foliage environments. In this thesis, the target recognition techniques based on RF signals have been extensively investigated. The following outcomes have been achieved. In order to tackle the issues related to target detection in foliage environments, a novel detection method based on DFS is proposed. Firstly, compared with several different types of RF signals, it has been found that the ultra-wideband (UWB) signal is the most appropriate one. Secondly, the statistical properties of UWB signal are selected as the extracted features. Finally, based on the collected data, the performance of the proposed methods has been extensively verified with different classification algorithms. To minimize the impact on detection accuracy due to unwanted clutters, a high-order cumulants (HOC) algorithm is proposed for feature extraction, which has a good immunity against background noise. First of all, 1-D diagonal slices of fourth-order cumulants are used to extract useful information from the received UWB signals. Then, different classification algorithms are used to demonstrate superior performance of this approach. Another critical issue related to the proposed method is the potential impact on detection accuracy due to weather variations. Thus, a novel method for recognition based on a hybrid differential evolution and flower pollination algorithm (DEFPA) is proposed. Using the collected data that has been gained from different weather conditions, the proposed DEFPA approach has not only improved the efficiency of the support vector machine (SVM) classifier, but also classification accuracy as multi-weather conditions are considered. Although the feasibility of using this technology in different weather conditions has been explored to some extent, we observe that it still cannot perform promisingly under severe weather conditions such as rain, fog, and snow. To address this problem, an Auto-Encoder/Decoder (Auto-ED) deep neural network is proposed. Experimental results demonstrate that the proposed approach can best tackle the challenge of target detection under severe weather condition in the foliage environment. At the end of this thesis, the summary of the overall work is presented and some possible future work is given.
Please use this identifier to cite or link to this item: