Research on Intelligent Grasping and Recognition Techniques in Robotics

This study removes accuracy constraints in the You Only Look Once version 3 (YOLOv3) network architecture to improve the robot’s ability to detect capture. Although YOLOv3 is widely used in the field of Object Detection, its accuracy is lower than Single Shot MultiBox Detector (SSD) algorithms in robot operation scenarios, especially in robotic manipulation contexts. first, there are three optimization measures that are offered to solve these problems: the model introduces The Edge function module, uses the Canny operator to extract boundary information as aprior knowledge, and combines Red Green Blue (RGB) and Edge data into a fourchannel input signal; second, it designs a dilation- Adaptive Spatial Feature Fusion network (ASFF) network structure to capture multiple images. scaling contextual information and expanding the sensory field without increasing the volume of computation; and third, using the Complete Intersection over Union Loss (CIoU) loss function allows for better consideration of the geometric relationship between the prediction field and the inscription field. The experiments show that the average accuracy of the improved new model increased by $2.98 \%$, which allows preserving the Iolov3 advantage in fast reasoning and improving accuracy and stability in performing tasks of capturing industrial robots.