大量收集之訊號的有效性確認，對於後續分析處理以達到現象探討及問題解析目的， 扮演了重要角色，本研究以復建工程中常擷取的慣性感測訊號為例，發展一訊號前處理 及篩選系統，以利使用者判別其有效及可用與否，減少巨量資料中的無效數據，及其混 淆問題探討的影響；此外，有助於非專業人員的後續數據使用，可避免錯誤判斷。 本研究之慣性感測訊號篩選系統分三部分： (1) 資料清理與整合：資料格式確認及漏點判斷，並以頻譜分析提出資料主要頻分量， 再資料整合為一平面檔(flat file)。 (2) 資料選擇與轉換：透過正常行走、在肩前舉、在肩外抬與肩外轉四種經常復健動作， 判斷動作共通頻率，選擇符合此頻率範圍之動作資料，將IMU 三軸主要頻率值繪於 三維空間，為後續資料挖掘之便利，使用座標轉換降低維度。 (3) 資料挖掘：復健動作為反覆且週期性，IMU 三軸主要頻率應相同，因環境雜訊影響 造成些微差異，故建立資料容忍範圍，收納可用資料。 建立篩選器後，研究中另外使用無線慣性感測器，先以上述動作設計模擬訊號，驗證系 統成效；再行檢視於北榮和三總收案之復健動作訊號，確認在臨床環境的有效性，並將 結果以混淆矩陣顯示。本研究所開發之系統除可快速篩選大量復健動作資料，也可提供 使用者立即判斷收錄資料之有效性，協助非工程人員、遠端復健之病患在復健療程資料 之運用。;In order to make sure the data can be used for subsequent analysis to achieve the purpose of phenomena discussion and problem analysis, it is important to confirm the validity of the collected data. This thesis developed a signal preprocessing and screening system to determine the availability of inertial sensing signals, which often retrieved in the rehabilitation-engineering. This screening system reduces the impact of bad data by reducing large amount of invalid data. In addition, the system helps non-professionals to use data and avoid false judgments. The function of the proposed screening system of this thesis can be divided into three parts: (1) Data Cleaning and Integration: Confirm the data format and determine the data loss, find the principal frequency by the spectrum analysis, and integrate the data into a flat file (2) Data Selection and Conversion: Input motion data can be filtered by 4 common rehabilitation actions, including walking, shoulder flexion, shoulder abduction and external rotation, which have the same frequency range. After spectrum analysis, the main three-axis principal frequencies values drawn in three-dimensional space. Coordinate transformations was used to reduce the dimensions to enhance the performance of data mining, (3) Data Mining: The three main IMU frequencies should be the same due to the repetitive and periodic characteristic of rehabilitation actions. Since slight differences caused by environmental noise, data tolerance range was established to collect available information. The function of the proposed screening system was validated by the simulation action data collected from wireless IMU. This paper also uses the signals of rehabilitation actions measured in the hospital to confirm the validity in the clinical setting. The result was displayed in a confusion matrix. The system developed in this study can not only rapidly screen a large number of rehabilitation action data, but also provide the user to quick judgment on the validity of the input data.
