5G+AI: Intelligent Sorting of Items in the Production Process 

In many industries such as industrial manufacturing and logistics, there are many needs for identifying, detecting and classifying products or objects. Traditional detection methods based on artificial vision are prone to fatigue, and cannot always guarantee high detection efficiency, and due to the physiological limit of the human eye itself, it is difficult to achieve high standards in terms of speed and accuracy. Therefore, the traditional manual detection method restricts the development and improvement of the productivity level and has been difficult to meet the needs of production. More and more sorting robots based on machine vision are widely used in various industrial production lines. That is, the target image information is obtained through the industrial camera, and the sorting of the grabbed target is completed. The traditional industrial robot visual sorting system uses classical computer vision theory, such as invariant distance, template matching, SURF feature and other algorithms to identify and locate workpieces, which has higher requirements on the scene environment, workpiece shape and background colour. The workpieces and objects on the on-site conveyor belt may have complex shapes and different postures and are densely placed together. In this case, the classical visual recognition and positioning algorithms cannot meet the requirements of high recognition rate and high precision.