论文部分内容阅读
为了解决岩石细观力学试验中图像处理过程复杂、质量不高及操作效率低等问题,将LS-SVM的分类方法与数字图像处理的阈值分割法相结合,提出了人机结合的岩石细观结构图像系统分析方法。该方法将图像分割问题转化为分类问题,通过对训练样本的学习,生成可将试验图像分类的LS-SVM分类机,从而提取岩石细观力学试验中得到的感兴趣区域的特征图像以及量化细观结构。对花岗岩图像进行处理,处理后的结果表明,该方法可以获得高质量的岩石细观图像处理结果,处理准确率达到96.82%。采用三步搜索法选取参数,能在保证图像处理质量的前提下提高参数选取速度;对训练样本进行稀疏化处理,可以提高分类效率,缩短分类时间;为了减小人为因素的影响,训练图像的选取应具有代表性,且在生成训练目标前需进行图像后处理。
In order to solve the problems of complex, poor quality and low operation efficiency in the meso-mechanics experiments of rock, the classification method of LS-SVM and the threshold segmentation method of digital image processing are combined to propose a man-machine combined rock meso-structure Image system analysis method. This method transforms the image segmentation problem into a classification problem. By learning the training samples, the LS-SVM classifier that can classify the test images is generated, so as to extract the feature images of the regions of interest in the meso-mechanics experiments and to quantify the details View structure. The results of processing the granite images show that this method can obtain high quality results of rock meso-images processing with the accuracy of 96.82%. Using three-step search method to select parameters can improve the speed of parameter selection while ensuring the quality of image processing. Thinning training samples can improve the classification efficiency and shorten the classification time. In order to reduce the impact of human factors, training images The selection should be representative, and the image post-processing needs to be done before generating training objectives.