Research On Shelf Commodity Detection Technology Based On Deep Learning

Supermarket has become an important option for the people's daily shopping due to its convenience.Commodity is the core of the supermarket.The state of the commodity directly reflects the running of supermarkets.In order to control the flow of commodities and provide consumers with a good shopping experience,supermarket managers often need to check the inventory of commodities on the shelves.However,the existing method of obtaining commodity information is mostly based on static software that provide store and query function of the type and amount of commodities,which requires a lot of people to provide auxiliary information.This way has many shortcomings,such as it has a large labor demand,and the information update is not in time.So there need a more intelligent way to obtain information of supermarket commodities in order to providing intuitional and timely commodities information.In view of the drawbacks of the above traditional commodities information management software,this paper studies the application of machine learning and compute vision in the detection of shelf commodities,and presents a new method for detecting the shelf commodities in a convenient and timely manner.In this paper,we firstly study the application of the state of the art object detection network Faster R-CNN and SSD to the detection of shelf commodities.Then,a new detection method of shelf commodities is put forward for bad performance of Faster R-CNN and SSD on the shelf images with many layers and many commodities.Firstly,the shelf image is divided by layer according to the vertical projection histogram of the Canny edge detection result.Then,the shelf image is divided according to the BRISK characteristic density distribution.Finally,the Faster R-CNN and the SSD object detection model are used to detect.By testing the image of the shelves in the real scene,the proposed method can get a detection accuracy of 94%in the seven categories of shampoo.