本篇論文主旨在於利用訊號轉換與機器學習建立一套依據腦波的專注力識別方法的研究。為建立此系統，本研究中請八個人擔任受試者，進行小測驗並擷取他們的腦波。我們運用經驗模態分解法（EMD）將原始的腦波轉換為八個本質模態函數（IMFs）以及一個趨勢項，將轉換後的數據運用快速傅立葉轉換（FFT）由時域訊號轉為頻域訊號，觀察其主頻率的分布。基於一般將多頻率的訊號疊加可能會導致一些特徵的消失，我們除了比較原始數據不同IMF的FFT訊號，也將多個不同FFT以串接的方式進行判別。我們將轉換、處理後的數據帶入機率類神經網絡（PNN）中，使其學習該訊號屬專注或是非專注，再比較不同訊號的組成，以及其所能達到的準確率。 本論文成功運用經驗模態分解法（EMD）、快速傅立葉轉換（FFT）以及機率類神經網絡（PNN），在專注與非專注腦波的辨識，平均有80%以上準確率。就結果而言，原始數據的FFT訊號有最高的單人以及平均準確率，而串接IMF1 ~ IMF3與IMF2 ~ IMF3的FFT訊號，其單人以及平均準確率雖並非最高，然而能夠達到75%準確率的人數卻為最多。 ;In this thesis, we build a system that could determine a brainwave belongs to attention or not. We invited eight people as our participants. They were asked to do a little test, and their brainwaves were captured at the same time. First, we used EMD to separate the raw brainwave into eight IMFs and a trend. By FFT, the signals were transferred from time-domain signal into frequency-domain signal. Since the superposition of different frequency-domain signals may cause some characteristics disappear, we also connected different FFT signals together instead of superimposed them. Last, we use PNN to learn the rules to achieve this system. We successfully used EMD, FFT and PNN to have great result in determining a brainwave belongs to attention or not, which has the average accuracy higher than 80%. In terms of results, the FFT signal of raw brainwave has the highest single and average accuracy, and FFT signal of IMF1 ~ IMF3 and IMF2 ~ IMF3 have the most participants can reach the accuracy higher than 75%.
