Biomimetic Underwater Vehicle Wall Surface Recognition Using Dynamic Pressure

本研究利用機器學習與分類理論探討仿生型水下載具游動時周圍的壓力場變化，並藉由實驗來作驗證。本實驗所使用之仿生型載具為長90公分、寬25公分、高50公分之類魚型水下載具，在載具兩側及頭部分別裝載壓力感測器，共七顆壓力感測器，用以感測游動時所產生的周圍壓力變化。 此論文中我們提出一個架構，將壓力感測器所量測到的信號經由濾波器及統計理論的分析過程，抽取出待分類的特徵，再利用支持向量機針對特徵向量來作分類。首先將訊號濾波，將濾波後的訊號作均方根，接著挑選出數種信號特徵作為特徵向量，爾後我們利用一個特徵選擇的方法，從所挑選的數種特徵向量適應性地選取較具代表性的特徵向量，最後再利用支持向量機來對不同類別的特徵向量予以分類。根據本文結論，仿生型載具能在自身游動時，憑藉著裝載的壓力感測器偵測周圍的壓力場變化，予以達到周圍偵測的效果，期許未來能進一步達成群游、避障等目的。 關鍵詞: 仿生型自主水下載具、側線系統、游動運動學、環境偵測Machine learning method is used to study the wall detection possibilities of a swimming Biomimetic Autonomous Underwater Vehicle (BAUV) by measuring the dynamic pressure surrounding the vehicle body. A BAUV of 90cm length, 25cm width and 50cm high with pressure sensors on body sides and on its head was built to sense the pressure around the BAUV. An architecture for the design of a two-class pressure signals classifier in an underwater environment is proposed. The classifier is based on a support vector machine, features vector are dynamic pressure signals processed by a band-pass filter and then their root-mean-square values are ensemble to represent the feature values. Feature vectors for classification are adaptively selected to form significant feature vectors by a feature selection method. Finally, the classifier is used for classifying the selected features. It is shown that a BAUV could use the dynamic pressure signals to estimate the presence of a solid wall beside it by the classifier. Keywords: Biomimetic Autonomous Underwater Vehicle, lateral line, swimming kinematics, environmental detectio