This paper addresses the autonomous operation requirements of robotic arm target grasping in smart manufacturing and flexible production scenarios, proposing a systematic approach that integrates vision-guided positioning with autonomous decision-making for target categories. Traditional robotic arm operation modes are constrained by single pre-set paths, making them unsuitable for dynamic multi-variety production environments. This research breaks through the limitations of existing spatial pose control by constructing a detection-classification-positioning synergistic decision-making framework, achieving closed-loop control of object recognition, classification screening, and precise positioning through multi-algorithm coupling. The system innovatively integrates a target classification module into the positioning process, focusing on three key aspects: hand-eye calibration parameter optimization, deep learning-based multi-scale target detection algorithms, and lightweight image classification models tailored for industrial scenarios. Experimental validation of specific object positioning demonstrates that this system can effectively realize target identification and autonomous robotic arm positioning in cluttered scenes, providing a robust and intelligent solution for dynamic industrial sorting and service robot applications.