基于Shearlet多尺度边界检测及融合的浮选气泡提取

Flotation Bubble Delineation Based on Shearlet Multiscale Boundary Detection and Fusion

针对浮选表面气泡图像边界弱、光照不均匀和气泡分布不均匀导致气泡提取困难的问题,提出了一种结合非下采样Shearlet变换(NSST)和多尺度边界检测及融合的浮选气泡提取方法。对气泡图像进行NSST分解,得到低频子带和多尺度多方向高频子带图像,通过构造自适应分数阶微分谷底检测模板提取低频子带的山谷边界,结合尺度相关系数及方向模极大值检测获取高频子带的边缘信息,再通过山脊特性判定从边缘信息中提取气泡的边界细节,最后进行多尺度边界融合、边界形态学处理以实现气泡提取。实验结果表明:该方法受噪声和光照的影响小,能有效提取出不同分布类型的气泡,其平均检测效率和准确率较现有方法有较大的提高,能够满足浮选工况动态变化的需求。

In order to solve the problems of weak edges, uneven illumination, and non-uniform bubble distribution in floatation surface bubble image, we propose a floatation bubble delineation method by combining nonsubsampled Shearlet transform （NSST） and multiscale boundary detection and fusion. First, we use NSST to decompose a bubble image into a low-frequency and multiscale multi-directional high-frequency subband images. Then, we use adaptive fractional order differential valley detection templates to extract valley boundary of the low-frequency subband, detect the edge information of high-frequency subband images based on scale correlation coefficient and direction modulus maximum, and extract boundary detail from the edge information by ridge characteristics determination. Finally, we implement bubble delineation through multiscale boundary fusion and morphological processing. Experimental results show that, the proposed method is affected little by noise and illumination, and can effectively delineate bubbles with different distribution types. The average detection efficiency and accuracy are much better than those of existing methods. This method meets the dynamic changes of flotation working condition well.