基于纹理去噪的摇床多矿带分割方法

Multiple Ore Zones Segmentation Method of Shaking Table Based on Texture Denoising

摇床矿带的颜色与分布直观地反馈了摇床的工况,生产现场普遍采用人工监控方式,劳动强度大,运用机器视觉方法可代替人工监控矿带,有效降低人力成本。为了解决摇床图像多矿带分割的难题,提出了一种基于纹理去噪-聚类分割的矿带分割方法。该方法使用直方图均衡化对输入灰度图像进行增强,提升矿带间对比度;引入双边滤波器(BTF)进行纹理平滑,避免平滑造成矿带边缘模糊;通过高斯混合模型(GMM)对纹理去噪后图像进行聚类分割,得到多矿带区域。还采用Dice系数、精确度和特异度作为评价指标,通过与直接分割法、其他去噪-分割组合方法对比,表明采用的方法能够实现多矿带分割,没有过分割与欠分割现象,指标明显优于其他方法,为摇床无人化生产奠定了基础。

The color and distribution of the ore zone of a shaking bed provides intuitive feedback on the working condition of the shaking table.Manual monitoring methods are generally used,which are laborintensive,and the use of machine vision methods can replace manual monitoring of the ore zone and effectively reduce labor costs.However,the low contrast,blurred edges and textural interference of the strip make it difficult to automatically monitor the strip based on the strip image.In order to achieve multiple mineral zones segmentation of shaking table images and solve the above processing difficulties,a mineral zone segmentation method based on texture denoising-clustering segmentation has been proposed.The method uses histogram equalization to enhance the input grayscale image which improve the contrast between mineral bands.With bilateral texture filter(BTF)for texture smoothing to remove texture interference and keep the edges of mineral bands,and performs cluster segmentation on the image after texture denoising by Gaussian mixture model(GMM)to obtain multiple mineral band regions.Dice coefficient,accuracy and specificity are used as evaluation indexes.By comparing with direct segmentation method and other combined denoising-segmentation methods,this method accurately achieves ore zones segmentation without over-segmentation and under-segmentation,and the indexes are significantly better than other methods.