Improved YOLO-V3 Workpiece Detection Method for Sorting

Abstract

Aiming at the problems of insufficient workpiece identification and detection accuracy in industrial scenes and difficulty in positioning multiple workpieces, a detection algorithm for identifying workpieces is proposed. Based on the YOLO-V3 algorithm, the network structure and multi-scale detection are improved, and the idea of hollow convolution is introduced. The residual dense block is used to replace the residual block in the original algorithm and combined with the convolutional layer to enhance the network's extraction of workpiece feature information. The original 3 scale detection is increased to 5 scale detection to improve the detection ability of small objects, Through the hollow convolution to expand the feature map of the workpiece to assist the network to extract deep-level object features, use the Kinect v2 sensor to collect the image of the workpiece and make it into a data set, the improved algorithm is tested on the data set, the experimental results show: The average detection accuracy of the workpiece reaches 92.98%, which is about 5% higher than the accuracy of the original algorithm. The combination of this method and robot grasping technology can replace manual labor to effectively complete the sorting of workpieces in industrial scenes.