Deep Learning Based K-Line Chart Recognition for Financial Quantitative Investment Analysis

In financial quantitative investment analysis field, most of existing artificial intelligence analysis methods are based on sequential data, and very few research apply expert intuitive image experience based intelligence method to analyze many K-line patterns. In this paper, a Faster R-CNN based image recognition method is proposed to analyze and predict the financial image (e.g., w-bottom patterns, pivots in entanglement theory, trend and consolidation patterns, etc.). The proposed method can intuitively recognize the K-line pattern images effectively and accurately. To do so, a Faster R-CNN is firstly constructed. The financial data images are then fed into the constructed neural network for training. What is more, the trained network model is used to predict some new financial images. The proposed method can not only recognize a financial K-line pattern feature, but also recognize multiple-features image patterns. The experimental results show that the recognition algorithm based on Faster -R-CNN greatly improves the accuracy of K-line pattern recognition, and the method is effective.