【作者】 唐立梅；

【导师】 沈永增；

【作者基本信息】 浙江工业大学 ， 控制理论与控制工程， 2001， 硕士

【摘要】 气液传质设备的研究和设计是化学工程的基本任务之一。复合塔是一种新型的气液传质设备，比较高的塔板效率是设计复合塔的基本要求，在建立复合塔板效率模型过程中，气液的比表面积a是一个需要获得的关键参数。通过照相分析的手段是获得这一参数较好的方法。 由于在化工现场采集到的原始图象质量较差，所以必须进行必要的图象增强。由于用传统的直方图均衡、滤波等方法处理的效果不理想，本文在模糊增强的理论基础上，提出了多级灰度增强的概念，利用模糊算子并通过定义隶属函数、饱和增强点和渡越点参数，给出了一种多级灰度增强的算法，该算法可根据要求对图象不同区域层次设计不同增强效果，使图象层次更加清晰。此算法简便易行，处理效果也比较好。 图象分割是图像处理以及模式识别过程中必须面对的问题，也是多年来这一领域的备受关注的焦点和难点。图像分割的方法灵活多样，理论背景差异也很大。模糊c-均值（FCM）算法是聚类分析理论及应用中的一个热点和比较成熟的算法。FCM算法用于图像分割，是一种非监督模糊聚类后标定的过程。由于FCM算法考虑到了每个样本点属于不同类型的程度，也就是类界限之间的模糊性，给分类提供了更多的选择机会。本文结合图象的一维直方图信息，定义了势直方图函数及剩余势函数，从而自动确定了分割类数，得到了比较满意的分割结果。 由于分割后的图象中，气泡目标彼此互相粘联的现象比较严重，本文采用二值形态学的方法进行分离。首先进行腐蚀，待目标完全分开后，计算目标的拓扑特征——欧拉数，然后在保持欧拉数不变的情况下，再对目标进行与腐蚀次数相同次数的膨胀，这样就把目标彼此完全分离。 在特征抽取阶段，以特征的可区别性、可靠性、独立性等原则，根据本文的分类目的，选取了气泡面积和圆形度两个特征构成特征向量。本文的分类属于类内鉴别，在此本文再次使用了模糊聚类算法，不仅进行了科学的分类，同时得到的聚类中心又是后序统计的关键参数。 最后本文对模糊聚类理论的发展进行了归纳总结。对FCM聚类算法中的加权指数m的选择、聚类中心的初始化方法以及聚类有效性进行了研究和探讨。更多还原

【Abstract】 The design of gas-liquid mass transfomi equipment is one of the basic task in chemical engineering. Compound tray is a kind of new tray in mass transform equipment. High mass transfer efficiency is the aim of compound tray design. In the course of compound tray efficiency modeling, gas-liquid interfacial areas a is a key parameter to be obtain. A good method to obtain is image analysis. Image enhancement is needed for the original images taken in industry locale are illegible. The enhance results obtained by conventional methods aren抰 ideal. Based on the fuzzy image enhancement theories, a multi-intensity enhancement algorithm is put forward with the definition of fuzzy operator. Various enhancement effects can be devised according to the processing requires of different image layer level. This algorithm is applicable for its briefness and good processing result. Image segmentation is not only an all-pervading problem but also a research focus and nodus in image processing and pattern recognition. Image segmentation algorithms are plenty in quantity and widely divergent in theory background. Fuzzy c-means (FCM) algorithm is a mature algorithm in clustering analysis field. FCM algorithm used in image segmentation is a course of unsupervised fuzzy clustering followed by demarcating. FCM algorithm offers advisable classification opportunities for the sample data because of its concerning of the boundary fuzziness among classes. Combining FCM algorithm with the histogram, we obtain good image segmentation results. Some bubble objects still conglutinate each other in the segmented image. Morphological-based algorithms are employed in the separating. In order to eliminate conglutination between any of the two objects, morphological erosion is operated to the image first. Then, morphological dilation is made the same times as erosion times on condition that keeping Euler number as a constant. We abstract characters on the principle of character抯 distinguishability, reliability and independency. Area and circularity are chosen as two elements of the eigenvector for they can remarkably characterize the interfacial areas. Classification become easy for the objects to be distinguished are included in one class. FCM algorithm is employed another time to obtain clustering center that act as key parameter in the statistic later. In the last chapter, Thorough discussions are made to the fuzzy c-means clustering algorithm. The selection of weight exponent m, the initialization of clustering center and the confinnation of clustering sorts are discussed.更多还原