Dense Object Detection Based On Physical Retail Scenarios

With the continuous development of generic object detection technology,the object detection capability for standardized pictures has been greatly improved.However,in many man-made specific scenes,objects are often irregular and difficult to detect.These scenes are often more meaningful for engineering applications.For example,the identification of densely arranged shelves of goods can greatly reduce the workload of goods consolidation for merchants.At the same time,under the economic background of the rise of the new physical retail industry,this article is also inspired,focusing the research direction on dense object detection based on the specific scenario of physical retail.It has been proved that accurate object detection in such dense scenes is still a challenging field.Even the most advanced object detectors cannot accurately locate densely aligned identical or similar detection objects.Therefore,this paper proposes a new object detection algorithm for this specific scene,which mainly includes the following two aspects of work:(1)Design and implementation of an Intersection over Union(IOU)sub-network to suppress a large number of overlaps or incorrect bounding box location that are prone to occur in dense scenes.(2)Introduce the attention mechanism on the basis of the model adding the IOU sub-network,and propose a dense object detection algorithm based on mixed attention to deeply mine the relevant information in the image to make the same or similar dense objects easy to merge or distinguish.Objects can be better separated from the background to further improve the detection effect of dense objects in physical retail scenarios.This article mainly conducts a series of comparative experiments with the existing mainstream object detection algorithms on the SKU-110 K data set that fully represents the physical retail scene.Experimental results show that the dense object detection algorithm based on physical retail scenarios proposed in this paper is superior to previous models.After adding the IOU sub-network and mixed attention module,the detection effect of shelf goods has been improved.These experimental results show the effectiveness of the proposed model.