基于图像分割和模式识别的钢材断口图像分析方法研究

发布时间：2019-05-29 11:45

【摘要】：落锤撕裂实验(DropWeightTearTest,落锤撕裂试验)是对预制缺口材料进行落锤击断的试验,通过计算断口表面的剪切百分数用来评估材料韧性。目前主要依靠人工目视判断韧性区面积百分比,主观因素影响精度,检测效率低下,迫切需要自动化检测仪器。但是落锤撕裂试件的断口的图像模式非常复杂:韧性区、脆性区混杂,高度起伏可达到30mm,对成像、照明尤其图像自动判别带来极大技术挑战。本文通过对落锤撕裂断口特征和机器视觉技术的深入研究,提出了基于图像分割和模式分类的评定方法,搭建了断口图像采集平台,开发了检测软件,开展了相关实验对整套检测系统进行了验证。首先,本课题结合断口表面的光学反射特性和三维特征,解决了断口图像的分析系统的总体设计;然后运用阈值分割,均值滤波,图像融合等算法对图像进行了预处理;针对不同的断裂类型图像进行图像分割处理;提取分割后区域的数字图像特征对高斯混合模型和支持向量机分类器进行训练,得到合适的图像分类模型,最终实现对断口图像的断裂区域的识别和分类。将算法的评定结果与人类专家评定的结果做对比,实验结果表明,本文设计的自动评定算法与人类专家的评定结果的绝对误差在4%以内,可以实现对落锤撕裂断口的自动评定。
[Abstract]:Drop hammer tear test (DropWeightTearTest, drop hammer tear test) is a test of falling hammer breaking of prefabricated notched materials. The shear percentage of fracture surface is calculated to evaluate the toughness of the material. At present, it mainly depends on artificial visual judgment of the area percentage of toughness zone, subjective factors affect the accuracy, detection efficiency is low, there is an urgent need for automatic testing instruments. However, the image mode of the fracture surface of the falling hammer tear specimen is very complex: the ductile zone, the brittle zone is mixed, and the height fluctuation can reach 30mm, which brings great technical challenge to the automatic discrimination of imaging and lighting, especially the image. In this paper, through the in-depth study of falling hammer tear fracture characteristics and machine vision technology, an evaluation method based on image segmentation and pattern classification is proposed, the fracture image acquisition platform is built, and the detection software is developed. The related experiments are carried out to verify the whole detection system. Firstly, combined with the optical reflection characteristics and three-dimensional characteristics of the fracture surface, the overall design of the fracture image analysis system is solved, and then the image preprocessing is carried out by using threshold segmentation, mean filtering, image fusion and other algorithms. Image segmentation is carried out for different fracture types of images. The digital image features of the segmented region are extracted to train the Gao Si hybrid model and the support vector machine classifier, and a suitable image classification model is obtained. Finally, the fracture region recognition and classification of the fracture image is realized. The evaluation results of the algorithm are compared with those of human experts. The experimental results show that the absolute error between the automatic evaluation algorithm designed in this paper and the evaluation results of human experts is less than 4%. The automatic evaluation of the tear fracture of the drop hammer can be realized.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】