This study aims to develop a prosthetic hand grip control system based on machine vision to improve the quality of life and self-care capacity of patients. In medicine and rehabilitation engineering, machine vision technology has been widely used to design intelligent prostheses to help patients restore limb function. Grip strength control is one of the key challenges in developing prosthetic hands; for example, patients need to appropriately control the grip strength of the prosthetic hand depending on the nature and size of the object to be gripped to prevent it from slipping or being damaged. This study combines machine learning and deep learning techniques to determine object grip information by analyzing images of such objects, including their type, texture, and size, so as to select the appropriate grip strength threshold. The electromyographic gesture-control mode is integrated with the visual recognition system to achieve active detection and control of the intelligent prosthetic hand. This research is also transplanted into the K210 main control board for offline recognition to achieve more efficient real-time performance. The experimental results demonstrate that the system achieves an object recognition accuracy rate of 90%, and the real-machine recognition rate is above 85%. The system successfully implements adaptive grasping for eggs (fragile items) and water bottles (rigid objects).