摘要
针对高炉内部复杂,影响煤气流分布因素多,且实时检测困难的问题,提出了一种基于感红外成像图像进行中心煤气流分布模式识别的方法.首先将红外图像进行灰度转化,利用边缘提取算法得出二值灰度点集,然后利用最小平方中值法进行椭圆拟合,并得出椭圆特征数据作为图像特征,最后利用递阶遗传算法将BP(back propagation)网络进行权值、阈值以及拓扑结构的优化,并用优化后的网络对特征数据进行模式识别,将中心煤气流的分布分为六类.研究结果表明,文中所提出的方法在训练时间和识别率上有较好的性能,识别率达到84.8%,可以满足对中心煤气流分布模式的检测.
A pattern-recognition method is proposed based on an infrared imaging image considering the complex in- ner environment of blast furnace, the quite large number of factors that affect the gas-flow distribution, and the difficulty in measuring the real-time gas-flow distribution. First, the infrared image is transformed into a grey-scale image, and an edge-extraction algorithm is used to obtain a binary grey point set. Then, the least median of squares is used to fit the ellipse and obtain the ellipse f~~ature data as an image feature. Finally, a hierarchical genetic algorithm is used to oplimize the BP nelwork weights, the threshnld, and the topological slrnelure. The oplimized nelwork performs pallern recognition of the characlerisiie data, and the cenler gas- flow distrihution is divided into six categories. The resohs indieatc that the proposed method yields belier per- formance in terms of the training time and achieves a high reognition rate of up to 84.8%, which satisfy the model test requirement of the center gas-flow distribution pallern detection.
出处
《信息与控制》
CSCD
北大核心
2014年第1期110-115,共6页
Information and Control
基金
国家自然科学基金资助项目(61164018)
内蒙古自然科学基金资助项目(2012MS0911)
关键词
特征提取
递阶遗传算法
BP神经网络
中心煤气流分布
feature extraction
hierarchical genetic algo-rithm
BP neural network
center gas-flow distribution
