Abstract: Due to complex and unstable operating environment of agricultural harvesting robot, robotic arm may be subject to external disturbances such as wind, vibration, or crop deformation, resulting in end effector shaking or mechanism deformation, which cannot effectively capture uncertain turning points of robotic arm's grasping end trajectory dynamically, thereby affecting accuracy of trajectory tracking．Therefore, a end effector trajectory tracking control method for agricultural harvesting robot robotic arm grasping based on CNN and visual image features was proposed．Trajectory information of agricultural harvesting robot robotic arm was extracted based on time series and spatial series, setting target forcing function, and spatiotemporal variation characteristics of robotic arm grasping end effector motion trajectory was extracted．CNN algorithm was used to calculate nonlinear error variation of dynamic parameters of harvesting robot robotic arm under nonlinear spring disturbance and environmental mechanical nonlinear disturbance．Kinematic orientation relationship between initial and end points was calculated, calibration error of visual image features was calculated, inputting calibration error into control actuator, visual calibration compensation feedback convergence function was used to adjust gap between target point and actual point, and to achieve trajectory control．Experimental results showed that proposed method accurately calculated dynamic parameter error of end effector trajectory of agricultural harvesting robot robotic arm grasping．Under visual direction calibration compensation for multi-point wave curves and multi-point rectangular curves, uncertain inflection points of trajectory could be captured dynamically and efficiently．Under nonlinear disturbance of internal springs and external environmental mechanics, proposed method performed well in tracking control grasping trajectory of robotic arm with good robustness．