YOLO物件偵測在臺灣股價指數期貨的應用和系統實作

A Trading System of Taiwan Stock Index Future by Using YOLO Object Detection Technology on Candlestick Charts

詹建昇(Ching-Sheng Chan)

指導教授 ： 范敏玄

國立臺中科技大學/資訊與流通學院/資訊管理系碩士班/碩士(2021年)

摘要

隨著資訊科技術的進步及網路環境的成熟，近年來人工智慧蓬勃發展，人工智慧已廣泛的應用在人臉識別、醫療影像診斷、自動駕駛、股價預測、影像物件偵測..等各種領域。本文透過程式進行標示(Label)K線(Candlestick)組合型態，應用深度學習取得價格趨勢轉折的K線(Candlestick)組合的圖形特徵，以此狀態預測未來走勢。同時應用YOLO物件偵測，彈性快速的建立一個簡單有效台指期當沖K線買賣模型和當沖即時K線(Candlestick)走勢監控系統，輔助投資人決策。

實驗結果顯示，台指期當沖K線物件偵測買賣模型的平均精度(mAP)為93%，測試勝率為52.52%，測試期間2019至2020年當沖報酬率為183.91%。實現了台指期當沖K線視覺化人工智慧看盤，應用YOLO物件偵測，輸出投資買賣訊號，供投資人台指期當沖操作輔助決策或程式交易參考。

關鍵字

機器學習 ； 物件偵測 ； YOLO ； 深度學習 ； 臺灣股價指數期貨 ； 當日沖銷

並列摘要

With the advancement of information technology and the maturity of the network environment, artificial intelligence has flourished in recent years. Artificial intelligence has been widely used in face recognition, medical imaging diagnosis, autonomous driving, stock price prediction, object detection, etc. field. In this study, uses a program to label candlestick charts and to find the feature in turning point of price trends on candlestick charts by using deep learning and predict future trends by this state. To create simple and effective candlestick charts trading model and candlestick trend monitoring system of Taiwan Stock Index Future by using YOLO object detection for investors to decision support or algorithmic trading.

The experimental results show that the mean average precision of the candlestick charts object detection model of the Taiwan Stock Index Future is 93%, the test was found to be successful by 52.52%, and a 183.91% profit was generated between 2019-2020.It realizes the visualized artificial intelligence of the candlestick charts of the Taiwan Stock Index Future and using YOLO object detection technology on candlestick charts for investors to decision support or algorithmic trading.

並列關鍵字

Machine Learning ； Object Detection ； YOLO ； Deep Learning ； Taiwan Stock Index Future ； Intra-day Trading

參考文獻

[1 ]SERDAR BIROGUL, GÜNAY TEMÜR,UTKU KOSE3 "YOLO Object Recognition Algorithm and Buy-Sell Decision Model Over2D Candlestick Charts" in IEEE Access, vol. 8, pp.91894-91915,May 2020"

[2] Alexey Bochkovskiy,Chien-Yao Wang,Hong-Yuan Mark Liao,"YOLOv4: Optimal Speed and Accuracy of Object Detection" pp.1-17,Apr 2020

[3] Omer BeratSezer,Ahmet MuratOzbayoglu,"Algorithmic financial trading with deep convolutional neuralnetworks:Time series toimage conversion approach" in ScienceDirect,pp.525-538, April 2018.

[4] Ross Girshick, Jeff Donahue, Trevor Darrell, Jitendra Malik,"Rich feature hierarchies for accurate object detection and semantic segmentation" in IEEE Access,pp.580-587,June 2014

[5] David M. Q. Nelson, Adriano C. M. Pereira, Renato A. de Oliveira,"Stock Market's Price Movement Prediction with LSTM Neural Networks.In Neural Networks" in IEEE Access,pp.1419-1426,2017