煤体CT图像阈值选取的准确性对于三维重建的模型能否还原真实煤体结构至关重要。Otsu阈值分割法（简称Otsu法）对煤体CT图像中孔裂隙结构的过分割，使得三维重建的煤体模型与实际不符。为克服该缺陷，首先探究了Otsu法对煤体CT图像阈值分割失效的机理，并据此确定合适的权重因子修正Otsu最佳阈值选择公式，提出了一种改进的Otsu阈值分割法（MP-Otsu阈值分割法，简称MP-Otsu法）。最后利用MatLab软件检测MP-Otsu法对煤体CT图像的二维分割效果，同时利用Avizo软件对煤体孔裂隙结构进行三维重建并对比改进前后孔裂隙参数的变化。结果表明：煤体内部矿物含量与Otsu法确定的阈值两者之间呈正相关关系。煤体内部矿物组分的存在导致目标和背景区域方差差异大，低孔隙率导致煤体CT图像灰度分布直方图呈单峰分布且不能提供足够的方差信息，这些特征是Otsu法对煤体CT图像阈值分割失效的主要原因。MP-Otsu法引入矿物含量与Otsu阈值的拟合曲线斜率，目标在图像中所占的大致比例等3个权重因子改进Otsu阈值分割法，得到的二值图像与原图像高度契合，能够准确提取孔裂隙目标区域。三维重建的模型孔隙率和最大喉道尺寸与Otsu法相比分别下降了96.18%和80.07%，有效克服了Otsu法对煤体CT图像过分割的缺陷。此研究能为进一步探究与孔裂隙结构相关的煤体物理特性提供基础。

The accuracy of CT image threshold selection is very important for the 3D reconstruction model to restore the real coal structure. Excessive segmentation of pores and fractures in coal CT images by Otsu results in the model structure not conforming to the reality. To overcome this defect, we first explored the failure mechanism of Otsu method on coal CT images, and then a modified method, called MP-Otsu, was present. Finally, Matlab software was used to detect the 2D segmentation effect of MP-Otsu method on CT images. Meanwhile, Avizo software was used to carry out 3D reconstruction of coal pores and fractures and compared the changes of pores and fractures parameters before and after improvement. Results showed that the minerals in coal and the threshold determined by Otsu showed significant positive correlation. The variance of target and background was very different due to the existence of internal mineral components. What’s more, low porosity contributed to the unimodal distribution of CT images’ grayscale distribution histogram and cannot provide enough variance information. All these features of coal were reasons that Otsu failed to segment the CT images. The MP-Otsu method introduced three weight factors such as the fitting curve slope of mineral content and Otsu threshold and approximate proportion of the target in the image, to improve the Otsu threshold segmentation method. The binarization images obtained were highly consistent with the original images and could accurately segment the target area of pores and fractures. Compared with Otsu method, the porosity and maximum throat size of 3D reconstruction model were reduced by 96.18% and 80.07% respectively, which effectively overcame the defect of Otsu’s over segmentation of coal CT images. This study may contribute to further exploring the physical characteristics of coal related to pores and fractures