基于U-Net和改进分水岭算法的露天矿爆堆矿石图像分割方法  

An Image Segmentation Method for Blast Pile Ore in Open-pit Mine Based on U-Net and Improved Watershed Algorithm

作　　者：阚玉达 KAN Yuda(China Railway 19th Bureau Group Corporation Limited,Beijing 100161,China)

机构地区：[1]中铁十九局集团矿业投资有限公司,北京100161

出　　处：《金属矿山》2023年第8期272-277,共6页Metal Mine

基　　金：国家自然科学基金项目(编号:42071453,41974028)。

摘　　要：爆破大块率是反映爆破效果的关键指标,其统计精度主要依赖于爆堆矿石图像分割的准确性。由于爆堆矿石图像中存在矿石目标分布密集、边缘对比度低等问题,致使传统图像分割方法难以准确分割爆堆矿石图像。因此,提出了一种基于U-Net和改进分水岭算法的露天矿爆堆矿石图像分割方法,以实现大块率的精准统计。首先利用无人机在哑巴岭露天矿爆破现场拍摄爆堆矿石图像,制作爆堆矿石图像数据集;然后利用深度学习算法建立了UNet网络架构,同时融合了高级语义信息和低级语义信息,建立了爆堆矿石图像分割模型,再利用训练后的模型对爆堆矿石图像进行初步分割,进一步采用基于距离运算的分水岭算法优化了分割结果;最后评估了该方法的分割精度。试验结果表明:该算法可准确分割露天矿爆堆矿石图像,为露天矿爆破大块率统计、爆破效果智能评价提供技术支持。The key indicator reflecting the blasting effect is the blasting block rate,and its statistical accuracy mainly depends on the accuracy of ore image segmentation in the blasting heap.Due to the problems of dense distribution of ore targets and low edge contrast in the ore image of the blasting heap,traditional image segmentation methods are difficult to accurately segment the blasting heap ore image.Therefore,this paper proposes a method based on U-Net and improved watershed algorithm for segmenting open-pit mine blasting heap ore images to achieve accurate statistics of the block rate.Firstly,unmanned aerial vehicles shoot the blasting heap ore images at the Yabaling Open-pit Mine,and create a blasting heap ore image dataset.Then,deep learning is used to establish the U-Net network architecture,and advanced semantic information and low-level semantic information are integrated to establish an ore image segmentation model for the blasting heap.The trained model is then used to perform preliminary segmentation on the blasting heap ore images,and further optimize the segmentation results using a watershed algorithm based on distance operations.Finally,the segmentation accuracy of the proposed method is evaluated.Experimental results show that this algorithm can accurately segment open-pit mine blasting heap ore images,providing technical support for open-pit mine blasting block rate statistics and intelligent evaluation of blasting effects.

关 键 词：U-Net网络 分水岭算法 爆堆矿石图像分割 爆破大块率 

分 类 号：TD235[矿业工程—矿井建设]