Stockmarket Pattern Detection Yolov8

🚀 YOLOv8s股票市场实时图案检测模型

该模型基于YOLO（You Only Look Once）框架，能够从屏幕捕获的股票市场交易数据中实时检测各种图表模式。它通过自动分析图表模式，为交易者和投资者提供及时的见解，辅助他们做出明智的决策。该模型在多样化的数据集上进行了微调，在实时交易场景中能够高精度地检测和分类股票市场模式。

🚀 快速开始

要开始使用YOLOv8s股票市场实时图案检测模型，你需要安装必要的库：

pip install mss==10.0.0 opencv-python==4.11.0.86 numpy ultralytics==8.3.94 openpyxl==3.1.5

💻 使用示例

基础用法

import os
import mss # type: ignore
import cv2
import numpy as np
import time
import glob
from ultralytics import YOLO
from openpyxl import Workbook

# Get the user's home directory
home_dir = os.path.expanduser("~")

# Define dynamic paths
save_path = os.path.join(home_dir, "yolo_detection")
screenshots_path = os.path.join(save_path, "screenshots")
detect_path = os.path.join(save_path, "runs", "detect")

# Ensure necessary directories exist
os.makedirs(screenshots_path, exist_ok=True)
os.makedirs(detect_path, exist_ok=True)

# Define pattern classes
classes = ['Head and shoulders bottom', 'Head and shoulders top', 'M_Head', 'StockLine', 'Triangle', 'W_Bottom']

# Load YOLOv8 model
model_path = "model.pt"
if not os.path.exists(model_path):
    raise FileNotFoundError(f"Model file not found: {model_path}")
model = YOLO(model_path)

# Define screen capture region
monitor = {"top": 0, "left": 683, "width": 683, "height": 768}

# Create an Excel file
excel_file = os.path.join(save_path, "classification_results.xlsx")
wb = Workbook()
ws = wb.active
ws.append(["Timestamp", "Predicted Image Path", "Label"])  # Headers

# Initialize video writer
video_path = os.path.join(save_path, "annotated_video.mp4")
fourcc = cv2.VideoWriter_fourcc(*"mp4v")
fps = 0.5  # Adjust frames per second as needed
video_writer = None

with mss.mss() as sct:
    start_time = time.time()
    last_capture_time = start_time  # Track the last capture time
    frame_count = 0
    
    while True:
        # Continuously capture the screen
        sct_img = sct.grab(monitor)
        img = np.array(sct_img)
        img = cv2.cvtColor(img, cv2.COLOR_BGRA2BGR)

        # Check if 60 seconds have passed since last YOLO prediction
        current_time = time.time()
        if current_time - last_capture_time >= 60:
            # Take screenshot for YOLO prediction
            timestamp = time.strftime("%Y-%m-%d %H:%M:%S")
            image_name = f"predicted_images_{timestamp}_{frame_count}.png"
            image_path = os.path.join(screenshots_path, image_name)
            cv2.imwrite(image_path, img)

            # Run YOLO model and get save directory
            results = model(image_path, save=True)
            predict_path = results[0].save_dir if results else None

            # Find the latest annotated image inside predict_path
            if predict_path and os.path.exists(predict_path):
                annotated_images = sorted(glob.glob(os.path.join(predict_path, "*.jpg")), key=os.path.getmtime, reverse=True)
                final_image_path = annotated_images[0] if annotated_images else image_path
            else:
                final_image_path = image_path  # Fallback to original image

            # Determine predicted label
            if results and results[0].boxes:
                class_indices = results[0].boxes.cls.tolist()
                predicted_label = classes[int(class_indices[0])]
            else:
                predicted_label = "No pattern detected"

            # Insert data into Excel (store path instead of image)
            ws.append([timestamp, final_image_path, predicted_label])

            # Read the image for video processing
            annotated_img = cv2.imread(final_image_path)
            if annotated_img is not None:
                # Add timestamp and label text to the image
                font = cv2.FONT_HERSHEY_SIMPLEX
                cv2.putText(annotated_img, f"{timestamp}", (10, 30), font, 0.7, (0, 255, 0), 2, cv2.LINE_AA)
                cv2.putText(annotated_img, f"{predicted_label}", (10, 60), font, 0.7, (0, 255, 255), 2, cv2.LINE_AA)
                
                # Initialize video writer if not already initialized
                if video_writer is None:
                    height, width, layers = annotated_img.shape
                    video_writer = cv2.VideoWriter(video_path, fourcc, fps, (width, height))
                
                video_writer.write(annotated_img)

            print(f"Frame {frame_count}: {final_image_path} -> {predicted_label}")
            frame_count += 1

            # Update the last capture time
            last_capture_time = current_time

        # Save the Excel file periodically
        wb.save(excel_file)

        # If you want to continuously display the screen, you can add this line
        cv2.imshow("Screen Capture", img)

        # Break if 'q' is pressed (you can exit the loop this way)
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break

# Release video writer
if video_writer is not None:
    video_writer.release()
    print(f"Video saved at {video_path}")

# Remove all files in screenshots directory
for file in os.scandir(screenshots_path):
    os.remove(file.path)
os.rmdir(screenshots_path)

print(f"Results saved to {excel_file}")

# Close OpenCV window
cv2.destroyAllWindows()

📚 详细文档

模型详情

模型描述

YOLOv8s股票市场图案检测模型能够实时检测股票市场屏幕截图中的关键图表模式。由于股票市场变化迅速，该模型的功能使用户能够及时获取信息，从而快速准确地做出明智决策。

该模型旨在处理股票市场交易图表的屏幕截图，能够检测诸如“头肩底”、“头肩顶”、“M头”、“股票线”、“三角形”和“W底”等模式。交易者可以优化他们的策略，自动做出交易决策，并实时响应市场趋势。

若要将此模型集成到实时交易系统或进行定制咨询，请通过info@foduu.com与我们联系。

支持的标签

['Head and shoulders bottom', 'Head and shoulders top', 'M_Head', 'StockLine', 'Triangle', 'W_Bottom']

用途

• 直接使用：该模型可用于对屏幕捕获的股票市场图表进行实时图案检测。它可以记录检测到的图案，对检测到的图像进行注释，将结果保存到Excel文件中，并生成一段时间内检测到的图案的视频。
• 下游使用：该模型的实时功能可用于自动化交易策略，为特定图案生成警报，并提高整体交易表现。
• 训练数据：该股票市场模型在一个自定义数据集上进行训练，该数据集包含9000张训练图像和800张验证图像。
• 适用范围外的使用：该模型不适用于与股票市场图案检测无关的目标检测任务或场景。

偏差、风险和局限性

• 性能可能会受到图表样式、屏幕分辨率和市场条件变化的影响。
• 快速的市场波动和交易数据中的噪声可能会影响准确性。
• 训练数据中未充分体现的特定市场模式可能会给检测带来挑战。

建议

用户应了解该模型的局限性和潜在偏差。在将模型用于实际交易决策之前，建议使用历史数据和实时市场条件进行测试和验证。

📄 许可证

文档中未提及相关许可证信息。

📞 模型联系信息

如有咨询和贡献需求，请通过info@foduu.com与我们联系。

引用信息

@ModelCard{
  author    = {Nehul Agrawal,
               Pranjal Singh Thakur, Priyal Mehta and Arjun Singh},
  title     = {YOLOv8s Stock Market Pattern Detection from Live Screen Capture},
  year      = {2023}
}