Research on Mechanical Arm Grinding Method Based on Improved CascadePSP Net

Abstract

Aiming at the problems of manual processing and grinding workpieces, such as time-consuming and low accuracy, an intelligent method for grinding workpieces was proposed. The dataset is preprocessed using image graying and guided filtering. The mechanical arm is used for machining and grinding, Adding semantic segmentation technology to realize accurate identification and location of machining trajectory, An improved CascadePSP Net is proposed to realize faster recognition while ensuring accuracy. By comparing the improved CascadePSP Net with the original network, the segmentation accuracy and training speed are improved. Use the Sober operator to extract the contour of the workpiece to be machined to determine the final machining path. The trajectory planning of the three-degree-of-freedom Dobot Magician manipulator is carried out by the fifth-order polynomial interpolation method and the Cartesian coordinate system method. Build an experimental platform for an image recognition robotic arm, and the comparison of the trajectory recognition method and the test experiment of the mechanical arm grinding system were carried out respectively. It verifies the feasibility of the proposed grinding method. The experimental results show that the method reduces the network training time, realizes high-efficiency and high-precision segmentation processing, thus improves the workpiece grinding efficiency and realizes the intelligent processing of workpiece batches.