%0 Journal Article
%T 基于YOLOv3的烧结机台车箅条图像角点定位<br>Corner Location of Sintering Machine Trolley Grate Image Based on YOLOv3
%A 廖婷婷
%A 李宗平
%A 曾小信
%A 欧阳攀
%A 李旭东
%J Mechanical Engineering and Technology
%P 205-212
%@ 2167-6623
%D 2023
%I Hans Publishing
%R 10.12677/MET.2023.122024
%X 准确抓取台车四个角点对于图像展平处理以及实现烧结机箅条智能监控具有重要意义，由于环境的复杂性，采用轮廓、特征点查找等方法很难实现角点的精准定位，而台车图像的高分辨率又给深度学习端到端模式的运行效率带来很大挑战。本文提出了一种深度学习与传统图像处理相结合的烧结机台车图像角点定位方法。首先从四个角点的大致位置提取出小块感兴趣区域，再采用YOLOv3检测模型，得到四个角点在感兴趣区域内的坐标，最后通过坐标换算得到角点的真实位置。通过定位角点可对图像进行畸变矫正，对计算箅条的斜率和间距具有重要意义。实际应用情况表明，该方法可实现快速精准的角点定位，并能适应生产场景，实际运行过程中达到了不低于95%的检测精度和少于1秒钟的检测时间，这为烧结机箅条的智能监控奠定了良好基础。<br />
Accurately grasping the four corners of the trolley is of great significance for image flattening pro-cessing and intelligent monitoring of sintering machine grates. Due to the complexity of the envi-ronment, it is difficult to achieve accurate positioning of corners by using contour, feature point search and other methods. The high resolution of the trolley image poses a great challenge to the ef-ficiency of the end-to-end mode of deep learning. This paper proposes a corner location method for sintering machine trolley image based on deep learning and traditional image processing: Firstly, extracting a small region from the approximate position of the four corners. Then, the YOLOv3 de-tection model is used to obtain the coordinates of the four corners in the small region. Finally, the real position of the corner is obtained by coordinate conversion. The distortion of the image can be corrected by the corner coordinates, which is of great significance for calculating the slope and spacing of the grate. The practical application shows that the method can realize fast and accurate corner location and adapt to the production scene. In the actual operation process, the detection accuracy is not less than 95% and the detection time is less than 1 second, which lays a good foun-dation for the intelligent monitoring of sintering machine grate.
%K 烧结机，深度学习，透视变换，目标检测；Sintering Machine
%K Deep Learning
%K Perspective Transformation
%K Object Detection
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=64885