基于Curvelet变换和压缩感知的煤岩识别方法

发布时间：2018-04-18 13:37

本文选题：曲波变换 + 煤岩识别　；参考：《煤炭学报》2017年05期

【摘要】：针对小波难以表达煤岩图像的边缘曲线特征,影响识别精度的问题,提出一种基于曲波变换的方法,对煤岩图像边缘进行稀疏表示。该方法通过曲波变换对煤岩图像进行曲波分解,得到各尺度层曲波系数,保留图像变换后的Coarse层低频系数,基于压缩感知理论,利用随机高斯矩阵对高频系数进行测量,实现高维系数降维,Coarse层低频系数与降维后的高频系数通过级联构成煤岩图像特征向量,最后结合支持向量机对煤岩图像进行分类识别。实验表明:通过曲波分解提取的特征能够有效地表达煤岩图像边缘的曲线特征,所提出方法煤岩的分类准确率达93.75%,比Haar小波方法提高了4.37%,所用降维方法比线性降维方法提取的特征向量更加有利于煤岩图像的分类识别。
[Abstract]:Aiming at the problem that wavelet is difficult to express the edge curve feature of coal and rock image and affect the recognition accuracy, a method based on Qu Bo transform is proposed to represent the edge of coal and rock image sparsely.This method decomposes the marching wave of coal and rock images by Qu Bo transform, obtains the Qu Bo coefficients of each scale layer, and preserves the low frequency coefficients of the Coarse layer after the image transformation. Based on the theory of compression perception, the high frequency coefficients are measured by using the random Gao Si matrix.The feature vectors of coal and rock images are constructed by cascading the low frequency coefficients of Coarse layer and the high frequency coefficients after dimension reduction. Finally, the classification and recognition of coal and rock images are carried out with support vector machine.The experimental results show that the features extracted by Qu Bo can effectively express the curve features of coal and rock images.The classification accuracy of the proposed method is 93.75, which is 4.37 higher than that of the Haar wavelet method. The feature vectors extracted by the reduced dimension method are more favorable to the classification and recognition of coal and rock images than the linear dimensionality reduction method.
【作者单位】：中国矿业大学(北京)机电与信息工程学院;
【基金】：国家重点研发计划资助项目(2016YFC0801800) 国家自然科学基金重点资助项目(51134024)
【分类号】：TD67;TP391.41

【相似文献】