圖型辨識在人工智慧領域已經行之有年，像是車牌、金屬表面瑕疵、人臉辨識或是植物辨識等運用，然而除了能在一般生活或在工業上的運用外，在金融上對於股票的應用大多以數值來做分析，以決策樹、技術指標及機器學習來做股價預測，投資者希望透過人工智慧找出過去股票規律性及漲跌的脈絡來預測獲利，K線分析是投資人常用的技術指標，它可以協投資人將過去發生狀況產生趨勢，投資人可以用日K線變化來推測明日的漲跌機會，許多投資人對於該項分析技術，紛紛投時間研究，成為投資股票的顯學。
隨者電腦科技的進步圖型辨識的運算速度及準確率也已經提升，本研究將運用過去 K線所產生股票連續圖型三個月的時間區間，分成連續兩日及連續三日兩種，透過卷積神經網路的辨識走勢的圖型，將辨識圖型再加以分成十類圖型作為日後對照趨勢使用，收集今日數值轉成本日連續 K線，分割對照組後，使用比對方式找出類似一日或兩的 K線圖，再找出隔日圖型，作為預測結果。
經驗證結果發現，只有兩成比例可以相符，有近四成比例是中二分之一開盤或收盤的趨勢，因此若無考量其的參數及天數，單一或兩天去預測隔日趨勢，失敗較高，一兩天趨勢容易隨者其他國家盤勢影響，因此辨識隔日走勢的需搭配隔日開盤做為參考提高勝率。
本研究未將成交量及其他新聞事件會影響股價之因子納入，只以過去發生，今日是否會再發生做本研究，並以圖像辨識為基礎討論，之後可將其他技術指標作為辨識或改成日K線來研究。

Pattern recognition has been used in the artificial intelligence (AI) field for years, for instance, license plate recognition, defect detection in metal surfaces, and biometrics. However, people believe pattern recognition can go further, especially in stock markets. Investors typically base their stock analysis on data tools, such as decision trees, technical indicators, and machine learning to estimate the regularity and trend. Since pattern recognition has become more accurate and powerful, more and more investors wonder whether this technique can enhance data tools’ strength, for example, candlestick patterns, a popular technical indicator. The experiments in this research aim to combine candlestick patterns with pattern recognition to increase the winning percentage.
First, this study applied Convolutional Neural Network (CNN) to recognize the candlestick pattern. The pattern is built according to three-month data from history and classified into two groups, one cycle for every two days and the other for every three days. After recognition, this paper divides the result into ten patterns for comparison and collection in the database. A new three-month candlestick pattern will occur by adding the latest data from today. After CNN recognizes the two groups, these patterns will be compared with the database. The most similar pattern for one or two days will reveal the prediction for tomorrow.
The finding reveals that only twenty percent match the prediction, and forty percent match the trend of either beginning or the end of the daily trade. Unless considering other parameters and variables, the prediction based on one or two days will not be accurate. According to the above conclusion, this paper suggests the prediction includes the opening trend of the next day to boost the win rate.
In Sum, the main contribution of this paper is using pattern recognition on the past three-month candlestick patterns to predict the stock performance on the next day. This work verifies the hypothesis of whether history tends to repeat itself. The process does not consider other indicators, such as trade volume or news, which might influence the stock market. Moreover, the technical indicator applied in the paper can change to others or reset for daily candlestick patterns for further research.

G.L. Morris, Candlestick Charting Explained： Timeless Techniques for Trading Stocks and Futures 2nd edition, McGraw-Hill Trade, 1995.
Hongduo Cao , Tiantian Lin, Ying Li , and Hanyu Zhang,2021.
Lee, C. H. L., & Liu, A. (2006, October). A Financial Decision Supporting System Based on Fuzzy Candlestick Patterns. In JCIS.
Wagner, G.S., & Matheny, B. L. (1993). Trading applications of Japanese candlestick charting (Vol. 38). John Wiley & Sons.
技術分析指標」懶人包！ - 股票技術面分析 ｜投資小學堂 (cmoney.tw) 。
李丞恩. (2020). 影線效果與 K 線技術分析的預測能力。
林呈勳. (2009). 結合技術指標與財經新聞之股票趨勢預測。
陳松男.(2010). 從理性投機的觀點來解釋隱含波動率偏斜：台灣股市指數選擇權市場的校準。
黃彤緹. (2019). 台灣股票技術分析投資績效實證研究： 以 KD 指標, MACD 指標, BBands 指標為例。
樓禎祺　&　何培基（2003）。股價移動平均線之理論與實證－以台灣股市模擬投資操作為例。育達研究叢刊。5_6，27-51。
鄭佩欣. (2009). 以 K 線型態辨識為基礎的投資決策模式。
謝劍平. (2014).現代投資學：分析與管理，智勝文化事業有限公司。
謝劍平. (2016).投資學：基本原理與實務，智勝文化事業有限公司。
盧繹帆. (2020). 燭台策略在台股的應用。