基于Gabor滤波器和潜在语义分析的烧成状态识别

摘要

A new method based on Gabor filter pre-processing and latent semantic analysis (LSA) was proposed to achieve accurate burning state recognition for the flame image of rotary kiln burning zone, with the goal of avoiding inaccurate feature extraction and poor burning state recognition result involved by the image segmentation-based methods. As the pre-processing step, the designed compact Gabor filter bank was employed to distinguish region of interests with distinguishable texture properties to facilitate the subsequent feature extraction and burning state recognition step. For the filtered flame images, the improved LSA method was employed to extract feature vectors to recognize the burning state, with the goal of reducing the dimensionality of features and avoiding zero-frequency problem. Experimental results showed the feasibility of this flame image-based method to extract feature directly to recognize the burning state. Moreover, the recognition performance was superior to the without Gabor filter-based method, traditional LSA-based method, temperature-based method, and image segmentation-based methods.%采用Gabor滤波器预处理与潜在语义分析相结合的方法,对回转窑烧成带火焰图像的烧成状态进行了更为准确的识别,避免了基于图像分割技术带来的不精确特征提取和较差的识别结果.基于所构建的压缩Gabor滤波器组对火焰图像进行预处理,增强具有不同纹理特性的特征区域的可分性以有利于后续的特征提取和状态识别步骤.对预处理后的火焰图像采用改进的潜在语义分析提取特征向量进行状态识别,以降低特征维数并避免零频问题.实验结果表明:直接从火焰图像中提取特征进行状态识别的方法是可行的,并且识别的效果较未采用Gabor滤波器预处理、传统潜在语义分析、烧成带温度和图像分割等方法的效果更优.