In the recycling factory, plastic bags can damage the screening equipment as they tangle around the discs, to prevent damage the screen, the machine must be stopped and the bags manually removed from the screen. It lead to the screen works at a low efficiency. Moreover, the identification of plastic bags is easy to duplicate because their color can match other objects in the garbage. So to overcome the case, this research develop an AI vision-based plastic bags detection algorithm that is capable of automatically picking out plastic bags of different colors in a waste stream. Firstly, a data collection is performed then these data is labeled and then feed into the YoloV9 architecture for training the model for recognition system. When the plastic bags moving on the conveyor, the system will used the trained model to classify the object and then calculate the real position of them. Then the position of the recognized bags are transmitted to the robot arm for picking process. The reliability, stability, and high successful rate of system are proving through the real experiment on Mk2 robot arm. It is also easy to install and maintain.