Lithology variation zones and adverse geology often cause serious disasters in tunnelling. It is necessary and important to develop non-destructive, in-situ, fast and intelligent means of identification methods for lithology variation zones and adverse geology, so as to take corresponding countermeasures in advance to reduce and prevent geological disasters. Here we report and demonstrate some recent progress on deep learning-based methods for intelligent lithology and adverse geology identification in tunnelling. These methods are based on in-situ data collection (non-lab environment), deep learning and fusion analysis of images, element and mineral data of rocks. The proposed methods were successfully applied in some tunnel engineering and can also be used in similar surface geological surveying and subsurface mining and logging analysis in underground projects.