由於機車在不同行駛條件下會產生不同之通過噪音，本研究試圖利用經驗模態分解方法（EMD）結合類神經網路技術應用在機車種類的辨識上，期望能分辨出不同機車在不同時速下之行駛狀況。在實驗數據取得方面，利用環保署的機動車輛通過噪音測試之方法，將聲音訊號錄製。在訊號處理及分類上，先利用經驗模態分解技術將聲音訊號藉由一組有限的內建模態函數來呈現。再將信號內部變化的時間尺度做為能量與頻率的直接析出，並計算各函數中的能量分布情形當作其特徵。在訊號分類上，利用倒傳遞類神經（BPNN）與廣義回歸類神經（GRNN）做訓練，進而連到車輛識別的目的，並比較兩種類神經的性能。從實驗結果證明EMD特徵擷取方法結合GRNN分類器有良好的車種識別能力。

In this paper, a scooter identification system using empirical mode decomposition (EMÐ) and artificial neural network is present. The experimental method using Environmental Protection Administration of the motor vehicle noise measurement method, and recoding out the pass-by noise signal, when scooter of different types at various speed. EMD is an adaptive method that can generate a set of intrinsic mode function (IMF) components to represent the original data, use of data within the time scale changes in the energy to do the direct analysis. Calculating the energy of each components distribution condition as characterized, combined with back-propagation neural network and general regression neural network for training, thus achieving the purpose of scooter identification and comparison of two neural network performances. The experimental result shows that EMD method combined with neural network classifier has a good identification rates, suitable to the characteristics of pass-by noise signal for the scooter identification methods.