隨著物聯網技術的成熟、人民環保意識抬頭，對於健康風險也越來越重視。穿戴式設備所產生的身體訊息資訊及環境監控的感測器(如空氣盒子)的數據也會爆炸性的成長，本研究將探討如何將這些數據做最有效的處理及儲存。本工作蒐集了全台灣的空氣品質感測器的數據，同時也應用長距離低功耗廣域網路技術，整合Arduino開發板，PMS5003T四合一環境感測器，LoRa模組，自行製作空氣品質感測器進行校園空品資訊的蒐集，獲得70%的傳輸成功率。此外，整合台灣將近3000個站點的數據於網頁平台做視覺化呈現。另一方面，採用樹莓派實做一邊緣計算架構，將感測器所蒐集到的數據利用訊息傳遞介面(Massage Passing Interface)在代表邊緣端的樹莓派做即時的資料預處理，不將所有的原始資料皆傳輸至雲端伺服器做處理和運算，藉此可以減少數據的傳輸量，進而降低能源的消耗。本工作也在樹莓派上實作物件辨識環境的建置，本研究結合了神經運算棒(Neural Ccompute Stick)提高樹莓派處理電腦視覺影像的能力，透過神經元運算棒的輔助，樹莓派在執行物件辨識程式時的每秒幀率獲得4倍的提升，同時也記錄樹莓派的能源消耗。 With the maturity of the Internet of Things technology and the rising awareness of the peoples environmental protection, more and more attention is paid to health risks. The body information generated by the wearable device and the data of the environmental monitoring sensor (such as the airbox) will also explode. This work will explore how to process and store the data most effectively. This work collects data of air quality sensors in Taiwan, and also applies Low-Power Wide-Area Network (LPWAN) technology, integrating Arduino development board, PMS5003T four-in-one environmental sensor, LoRa module to make own sensors. The self-made air quality sensor is used to collect the campus air quality information and achieved a transmission success rate of 70\%. In addition, the integration of data from nearly 3,000 sites in Taiwan is visualized on the web platform. On the other hand, using Raspberry Pi implement an edge computing architecture, the data collected by the sensor is processed by the Message Passing Interface (MPI) on the edge of the Raspberry Pi. Instead of all original data is transmitted to the cloud server for processing and calculation, by reducing the amount of data transmission, thereby reducing energy consumption. This work is also used to implement the object detection environment on the Raspberry Pi. This study combined the Neural Compute Stick (NCS) to enhance the ability to process computer vision images on Raspberry Pi. Through the aid of NCS, the Raspberry Pis frames per second (FPS) is increased by 4 times when the object detection program is executed, and the energy consumption of the Raspberry Pi is also recorded.
