受煤矿井运输巷道空间限制，将钻孔机器人设计为灵活布局的钻进和钻杆加卸2个独立系统，利用基于合作靶标的单目视觉识别，将钻进和钻杆加卸2个系统关联，解决两系统递送钻杆的协同作业问题。然而，煤矿井下钻孔施工中场景复杂，存在孔口煤渣、污水污染合作靶标、粉尘浓度大能见度低、人造光源亮度低、光照不均匀使靶标图像对比度低，以及矿灯直射合作靶标反光等问题，会影响合作靶标图像清晰度，使基于靶标识别加卸钻杆的位置的精度降低，进而影响钻孔机器人自动加卸钻杆的成功率。因此，首先针对施工中孔口返出的煤渣、煤泥污染合作靶标的问题，利用形态学滤波中先腐蚀，再膨胀的开运算，清除煤渣间断斑块和突出靶标像素点；先膨胀再腐蚀的闭运算，消除小的孔洞和间断细小鸿沟，达到消除固体污染物对合作靶标的影响；其次针对粉尘、岩尘、水雾和污水影响图像对比度问题，通过暗通道先验求出初始透射率，采用梯度域导向滤波精细化透射率，利用导向滤波得到改进暗通道，并求大气光值，利用大气散射模型得到除雾后的图像，实现清晰化图像、增强图像对比度目的；最后对亚像素级角点和方向进行精细化，优化能量函数，使用种子棋盘格沿4个边缘生长棋盘格，用能量变化情况确认是否生长出最大的棋盘，补全因强光反射缺失的合作靶标图像。实验结果表明：将形态学滤波、去雾算法和棋盘格生长等方法相融合，可增强煤矿井下复杂场景钻孔施工中合作靶标图像的清晰度，提高基于视觉识别的图像精度，为钻孔机器人自动加卸钻杆的精确控制奠定基础。

Due to the space limitation of underground roadway in coal mine, the drilling robot is designed with two independent systems, i.e., the drilling system and the drilling pipe loading and unloading system, which are in flexible layout. Using the monocular visual identification based on cooperative target, the drilling system is associated with the drilling pipe loading and unloading system for the cooperative operation of the two systems to deliver the drill pipe. However, there are problems such as borehole cinder, cooperative target polluted by sewage, high dust concentration, low visibility, low luminance of artificial light source, uneven illumination resulting in low contrast of target image, and the reflection of cooperative target under the direct lighting of mining lamp in the complex scenarios of drilling construction in the underground coal mine, which will affect the resolution of cooperative target image and reduce the accuracy of target identification and the position to load and unload the drill pipe, thus influencing the success rate of automatic loading and unloading of drill pipe by drilling robot. Therefore, in view of the problem of cooperative target pollution by coal cinder and coal slime returned from the borehole in construction, the open operation of first corrosion and then expansion in morphological filtering was used to remove the discontinuous patches and prominent target pixels of coal cinder. Meanwhile, the closing operation of first expansion and then corrosion was also adopted to remove the small holes and intermittent small gaps, thereby eliminating the impact of solid pollutants on the cooperative targets. Besides, to eliminate the effect of dust, rock dust, water mist and sewage on image contrast, the initial transmittance was obtained by dark channel prior, the transmittance was refined by gradient domain guided filtering, the dark channel was improved by the guided filtering, and the atmospheric light value was solved. On this basis, the defogged image was obtained with the atmospheric scattering model, so as to realize the purpose of sharpening the image and enhancing the image contrast. Finally, sub-pixel corners and directions were refined, and the energy function was optimized. In addition, the seed checkerboard was used to grow checkerboard along the four edges, and the energy changes were used to confirm whether the largest checkerboard was grown, and thus the missing cooperative target image due to strong light reflection was completed. The experimental results show that the combination of morphological filtering, defogging algorithm and growth based checkerboard corner detection method could enhance the resolution of the cooperative target image in the complex scenarios of drilling construction in underground coal mine and improve the image accuracy based on visual identification, laying a foundation for the precise control of automatic loading and unloading of drill pipe by the drilling robot.