自組織地圖之Wi-Fi室內定位系統降噪

Noise Reduction by Self-Organizing Map for Wi-Fi Indoor Positioning System

黃宥翔(Yu-Hsiang Huang)

指導教授 ： 劉宏煥

中原大學/電機資訊學院/電子工程研究所/碩士(2019年)

https://doi.org/10.6840/cycu201900656

摘要

隨著科技的進步，定位技術也愈來愈被廣泛應用於我們的生活當中，為了實現定位技術就需要收集大量數據，並且通常會存在著干擾定位的噪音，也因此容易讓訓練出來的模型準確率不佳，本文使用Wi-Fi訊號來進行實驗，因為Wi-Fi裝置成本低、獲取容易、架設門檻低，同時在現代幾乎人人都有一台智慧手機，因此大大增加的Wi-Fi訊號的重要性，在本實驗中我們對Wi-Fi訊號中的Sample，BSSID和RSSI進行數據預處理，然後將其導入混合模型進行訓練分析。混合模型包括自組織映射（SOM）和人工神經網絡（ANN），我們首先使用SOM模型對數據進行分群並利用MID與SOM收斂的特性找到數據的重疊點，再來通過ANN模型訓練分析這些重疊點找出會干擾準確率的噪音並且刪除，最後使用噪音去除前後的兩個數據集對ANN模型進行訓練，並將一個新採集的Wi-Fi資料來當作測試集，進而比較兩個ANN模型的預測準確率來驗證本文所提出的方法。

關鍵字

自組織映射神經網路 ； 人工神經網路 ； 混合模型

並列摘要

With the advancement of technology, positioning technology is also increasingly used in our lives. In order to achieve positioning technology, a large amount of data needs to be collected, and there is often noise that interferes with positioning , which is likely to cause poor accuracy. This paper uses Wi-Fi signals because Wi-Fi devices are low in cost, easy to obtain, and have low barriers to building. At the same time, almost everyone has a smart phone in modern times, so the importance of greatly increasing the Wi-Fi signal. In this experiment, we perform data preprocessing on Sample, BSSID and RSSI in Wi-Fi signals and then import it into the hybrid model for training analysis. The hybrid model includes the Self-organizing map (SOM) and the Artificial Neural Network (ANN). We first use the SOM model to group the data and use the characteristics of MID and SOM convergence to find the overlap of the data, and then use the ANN model to train and analyze these overlapping points to find the noise that will interfere with the accuracy and delete. Finally, the ANN model is trained by two data sets before and after noise removal, and a newly acquired Wi-Fi data is taken as a test set, and then the prediction accuracy of the two ANN models is compared to verify the proposed method.

並列關鍵字

Self-organizing map ； Artificial Neural Network ； Noise Reduction

參考文獻

[1] C. Liu, "Implementation of Quick Wi-Fi Radio Fingerprint Collection on Android Smartphone for Indoor Positioning Systems," M.A. thesis, Chung Yuan Christian University, pp. 1-13, July 2018.

[2] T. Kohonen, S. Kaski, P. Somervuo, K. Lagus, M. Oja, and V. Paatero, "Self-organizing map," BIENNIAL REPORT 2002-2003, Chapter.8, pp. 113-122, February 2004.

[3] Y. LeCuni, L. Bottoui, G.B. Orr, and K.R Miuller, "Efficient BackProp," Neural Networks: Tricks of the Trade, pp. 59-50, 1998.

[4] B. Widrow, and M.A. Lehr, "30 years of adaptive neural networks: perceptron, Madaline, and backpropagation," Proc. IEEE, Volume 78, Issue 9, pp. 1415-1442, September 1990.

[5] K. Eremenko, "Deep Learning A-Z™: Artificial Neural Networks (ANN)," August 2018.