基于改进聚类算法的烧结火焰图像分类

Classification of Sintered Flame Images Based on Improved Clustering Algorithm

烧结机尾断面火焰图像能够最直接有效地反映烧结终点的状态。充分利用火焰断面图像所蕴含的有效信息对烧结终点状态进行分类具有可行性及工程实际意义。提出一种K均值结合图像颜色特征的分类算法,实现对烧结机尾断面火焰烧结状态的分类。首先,对90张火焰图像进行预处理,在烧结机采集的320 m2断面图像上按分辨率3024×1700像素对红火区域进行统一裁剪,提取烧结核心区域。对裁剪图像进行K均值分割,并对K分别为2,3,4的分割图像进行比较,结果表明K为3时的分割结果可以较准确地将火焰的红火区分割出来。其次,由于分割后的图像仍存在其他非红火区域,为了准确地提取红火区的几何特征,进一步对红火区进行颜色特征提取,得到最终的红火目标区域分割图像。最后,将提取的目标图像几何特征作为数据集,采用fuzzy C-means(FCM)算法对烧结终点状态进行分类。与传统FCM算法的分类结果对比表明,所提火焰图像分类算法改善了分类效果。

The flame image at the tail section of the sintering machine can reflect the state of the sintering endpoint directly and effectively.It is feasible and practical in engineering to utilize the effective information in the flame image to classify the state of the sintering endpoint.Therefore,this paper proposes a classification algorithm based on K-means with the image color features to classify the sintering states of the flame at the tail section of the sintering machine.First,90 flame images were preprocessed.The section images with 320 m2 that were collected by the sintering machine were cut uniformly in the red fire area according to the resolution of 3024×1700 pixels.Then,the core areas were extracted and sintered.The K-mean segmentation of the clipped image and the comparison of the segmentation images with K values of 2,3,and 4 show that the segmentation results when K=3 can be used to segment the red fire area of the flame more accurately.Second,the color features of the red fire area are further extracted to obtain the final red fire target area segmentation image,since there are still other nonred fire areas in the segmented image.Finally,the geometric features of the extracted target image were taken as the dataset,and a fuzzy C-means(FCM)algorithm was used to classify the sintering end state.The classification effect of the proposed flame image classification method improves more than that of the traditional FCM algorithm.