Abstract: In CNC machining of critical aero-engine parts, the misuse of excessively worn cutters causes scrapping of key components and sharply increases production costs. Therefore, fast and accurate monitoring of tool wear is required. To address problems of small detection range and unclear surface morphology in current in-situ tool wear detection methods, a novel machine vision-based on-machine detection system for milling cutter flank wear was developed. Embedded devices are used to acquire clear images of cutter flank morphology over a large field of view. Firstly, the reflective characteristics of tools are analyzed, and a novel illumination source is designed to achieve clear and uniform lighting of the cutter flank. Secondly, combined with a high-resolution CMOS camera and a telecentric lens, high-quality images that accurately reflect the cutter surface morphology are obtained. Finally, for wear region extraction, the segment anything model (SAM) with strong generalization and high segmentation accuracy is adopted and improved to better detect details of milling cutter wear regions. Compared with measurements by extended focus microscope, the maximum measurement error of the system is less than 20 μm, showing that it can effectively monitor cutter wear during machining.