Research On Commodity Recognition Algorithm For Unmanned Vending Machines Based On Linu

In recent years,with the rapid development of artificial intelligence,the unmanned retail industry has begun to rise.Among them,unmanned vending machines have become a research hotspot in the offline retail industry.Target detection in deep learning,a new technology,can identify goods more accurately and can be applied to a wider range of scenarios.The introduction of these new technologies is expected to promote the further development and popularization of the unmanned vending machine industry.In this thesis,the commodity identification algorithm is studied under the application scenario based on unmanned vending machine.The main work is as follows:(1)Research the current mainstream commodity identification algorithm for target detection,test the detection accuracy and speed of the current mainstream target detection algorithm through the same data set MS COCO2017,and select the most suitable commodity identification algorithm according to the actual application scenarios of unmanned vending machines for in-depth research.(2)The fisheye camera is used to collect the commodity identification data set,and the chessboard calibration algorithm is used to optimize the image correction for the collected data set.After the completion,Label Img software is used to mark the data set to meet the requirements of algorithm model training.(3)Through the previous analysis and comparison and algorithm test,YOLOv5 algorithm is selected for further study.Subsequently,based on the YOLOv5 s target detection algorithm,this thesis improves it and designs a lightweight commodity identification algorithm YOLOv5 sF.Compared with the original algorithm,this algorithm greatly reduces the volume of the model and improves the reasoning speed without reducing the detection accuracy.At the same time,the model transformation,the quantization of Int8 and the deployment of hardware platform of YOLOv5s-F algorithm are completed,and its actual performance is finally tested.(4)According to the actual needs and application scenarios,the overall application scheme of the unmanned vending machine is designed,including hardware design and software design.The hardware design includes the design of face-brushing payment module,product identification module and video monitoring module,and the software design includes Linux platform construction,Android operating system porting,openwrt operating system porting,face-brushing payment program design,product identification algorithm design,GUI interface design and so on.Finally,the YOLOv5s-F algorithm designed in this thesis can identify goods more quickly and accurately,which improves the recognition accuracy by 5% compared with the traditional YOLOv5 s.And the unmanned vending machine scheme given in this thesis can complete the functions of commodity target detection,face payment,remote unlocking and remote monitoring,etc.,and has broad market application prospects.