本研究應用電腦視覺技術OpenCV（Open Source Computer Vision）並結合機器學習中的集群分析K-means提出一個可以快速計算極火蝦數量的智慧數蝦系統。針對企業需求，所以本系統不僅要考慮執行效率與準確度，同時還需要注意系統成本與彈性，基於彈性及成本的考量將計算蝦隻的影像處理實作於AWS Lambda，透過REST（REpresentational State Transfer）Web Service的呼叫方式來存取，這樣的架構可以在請求數量暴增或是影像處理優化的情況下，均可無縫的進行更新，且把主要計算置於雲端就可以按照使用量來計費，且減少終端裝置樹莓派的成本。本研究在數蝦影像處理的流程依序是1.取得極火蝦影像的HSV（Hue Saturation Value）色彩空間特徵值，2.根據該影像特徵值過濾目標影像，3.進行二值化，4.影像分割找出輪廓，根據閥值找出極火蝦的候選區域，計算極火蝦的候選區域的面積，5.透過分群決定蝦隻面積，6.最後計算蝦隻數量。透過實驗結果顯示在計算150隻以下的極火蝦，影像處理的部分可以在0.1秒內計算完畢，且準確率高達95%以上。

This research adopts computer vision technology, OpenCV (Open Source Computer Vision), combined with K-means in machine learning to propose an intelligent shrimp counting system that can quickly calculate the number of the shrimps. In the case of considering the higher flexibility and lower cost of this application, we introduce the emerging technologies of serverless and IoT (Internet of Thing) such as AWS (Amazon Web Service) Lambda and Raspberry Pi. This research is, therefore, mainly designed to apply the computer vision to undertake the counting of shrimps automatically. OpenCV provides plenty of computer vision applications and often cooperates with the Raspberry Pi and AWS Lambda. The steps of image processing for accurately counting the shrimps are as follows: (1) capture the image, (2) filter and remain the sampling color, (3) threshold the filtered image, (4) contour the blobs in the image (5) determine the area of one shrimp (6) count the number of the shrimps. Concerning the performance and flexibility, we embed the image process into AWS Lambda function. Experimental results of counting shrimps (Neocaridina heteropoda var. red) show that our proposed application completes the counting operation of 150 shrimps in 0.1 second and the accuracy is up to 95%.