The invention discloses an image feature based intelligent wild lithology identification method. The method includes the following steps: firstly, for image importing and editing, importing rock mass, rock or rock core images to form a rock mass image library, renaming images in the rock mass image library, and adding attributes; secondly, for image learning, applying a digital image processing algorithm to learn images in the image library; thirdly, for image searching, uploading images needing identification, setting image identification fuzzy degree parameters, searching, comparing and selecting the uploaded images and the images in the rock mass image library; fourthly, for image comparison and screening, displaying a plurality of images after searching, comparing and selecting, performing primary screening, selecting single images and displaying attributes of lithology to obtain similar images, subjecting search results to secondary screening, and finally determining lithology of the rock mass in the images needing identification. The problems of low accuracy and efficiency in conventional lithology identification operation are solved, and the method can be widely applied to the field of geological digital image processing.