Create a one-channel cv::mat object from 2d cv::mat object

[UnlimitedR]

they are later feed to libtorch for machine learning usage, which requires a 3D data. but I find I can directly reshape the tensor