- 學位論文
調適性網路模糊推論於水下聲源訊號辨識系統之研究
The Research of Adaptive Network Fuzzy Inference System Applied on Underwater Acoustic Signal Recognition
摘要
水下聲源信號由於受到各種因素的影響，具有非線性及時變的特性。因此本論文針對水下聲源訊號提出實際可行的理論與處理方法，一是利用希爾伯特轉換取得水下聲源訊號之特徵參數;二是調適性模糊推論系統(ANFIS)的建立。最後對此兩部分進行整合，架構一確實可行的識別系統。 進行特徵參數抽取分析時，就訊號特性分析及特徵參數之求取分別做研討。經驗證，使用具多頻解析特性之小波封包分解法，而後再利用希爾伯特轉換，以獲得各個船隻具代表性的樣板特徵參數。 調適性模糊推論系統建立時，利用各樣本類別的特徵參數，經由規則及歸屬函數的建立程序，建立初始的調適性模糊推論系統，並使用混合式學習演算法進行求解。然後，可依所求解的狀況增減模糊規則數及歸屬函數，達成最佳化的辨識系統。
並列摘要
Underwater acoustic signal is affected by various factors, and it reveal characteristics of non-linear and time-variant. Therefore, a practical recognition system is proposed which consist of two parts. The One is underwater acoustic signal feature extraction by using wavelet packets and Hilbert Transform. The other is the signal pattern recognition by using Adaptive Network Fuzzy Inference System (ANFIS). Finally, combine the two procedures and establish a practical recognition system. During the feature parameter extraction stage, signal characteristic analysis and feature selection is discussed. It has been proved that using the wavelet packet decomposition method, and then apply the Hilbert transform can get representative pattern feature parameters of each sample classification individually. During the Adaptive Network Fuzzy Inference System modeling stage, each ship’s template feature parameters are utilized on to construct the preliminary fuzzy rules and membership functions, and solve the result by using the hybrid learning algorithm. Then, based on the adjustment of fuzzy rules or membership functions, an optimum recognition system is obtained.