基于深度学习的罐式炼炉送料口视觉检测与跟踪方法

Visual Inspection and Tracking Method for Feed Port of Tank Furnace Based on Deep Learning

通过计算机视觉技术实现送料口的准确检测及跟踪成为罐式炼炉自动控制的关键。但传统视觉检测和跟踪算法受炼炉高温产生烟雾的影响,易导致送料口检测和跟踪失效,为了更好地检测与跟踪视频流中送料口,提出了一种基于Yolo v3深度学习的检测及跟踪方法。基于自建的大量有烟雾干扰的图片样本,通过使用K-means聚类方法对数据集进行聚类分析,对送料口的检测与跟踪展开研究。实验结果表明:改进后的Yolo v3算法对实验目标准确率达到97.20%,检测速率达到了30帧/s,能实时检测和跟踪实际情况下送料口的位置信息,且具有较高的准确性和鲁棒性。其检测准确率和检测速率相比传统目标检测方法有所提高,准确性和实时性均满足自动控制生产要求,说明将深度学习技术应用到工业生产有着巨大的应用前景。

Due to the change in the load of the furnace,the rotation speed of the tank furnace is not uniform,and the traditional industrial automatic control method is difficult to achieve accurate feeding.The accurate detection and tracking of the feed port through computer vision technology has become the key to the automatic control of the tank furnace.However,the traditional visual detection and tracking algorithm is affected by the high temperature generated by the furnace,which is easy to cause the detection and tracking failure of the feed port.In order to accurately and robustly detect and track the feed port in the video stream,this paper proposes a deep learning detection and tracking methods based on Yolo v3.Based on the self-built large sample of smoke interference images,the Kmeans clustering method is used to cluster the dataset,and the detection and tracking of the feeding port are studied.The experimental results show that the improved Yolo v3 algorithm achieves an accuracy of 97.20%for the experimental target and a detection rate of 30 f/s.It can detect and track the position information of the feed port under real conditions in real time,and has high accuracy and robustness.The detection accuracy and detection rate are improved compared with the traditional target detection method.The accuracy and real-time performance meet the requirements of automatic control production,indicating that the application of deep learning technology to the industry has great research prospects.