Research And Application Of FMCG Detection Based On Slant Object Detection Technology

Object detection has always been a hot topic in academic and industrial research,with the rapid development of deep learning in recent years,but there are still many problems in the practical application which has an influence on the accuracy rate.the objects are tiny,tremendous,and dense typically tilted arouse a a problem about how to overcome these negative factors while improving accuracy and detection speed.Have produced many excellent advanced research modle in recent years,such as Yolo,Retinanet,Faster-Rcnn.However,these methods are only effective for sparse object scenes but for intensive object detection tasks with tilted objects,because using a rectangular candidate box in a tilted scene will cause adjacent objects to overlap.The overlapping area is so large that the excessive interference characteristics is appeared during training,and the threshold is not well selected when non-maximum suppression is performed.In order to solve the above problems,this paper proposes a new type specifically for this.a rectangularObject with a tilted angle as a candidate Object directional match network(DMnet),which can effectively avoid the interference of adjacent features and the problem of non-maximum value suppression.Since the inclined rectangular Object is used as the candidate Object,the subsequent IOU calculation,the regression of the tilted box,the Loss function,etc.need to be improved.In this paper,the corresponding solutions are given for the above problems,and on the FMCG data set.Experiments were carried out to verify the feasibility of the proposed method and achieved good results.The mAP value on the proposed method is higher that the method of positive rectangular box as a candidate Object of object detection algorithm.In this paper,the above proposed method is also applied to the FMCG System.Beer manufacturers need to conduct statistical analysis on beer sales in the market.Therefore,it is necessary to keep abreast of the sales of various types of beer in the supermarket.For statistical purposes,it is very time-consuming and laborious.Now using object detection and recognition technology to automate this work,the FMCG identification system can automatically change the type and quantity of beer in the picture,and the changed labelImg is able to Label the tilt object.The taining data is pre-processed to remove the bluri mages.This system is a Web service built usingthe Flask Objectwork,and uses TensorFlow Serving as a model management tool,Due to some rules for such scenarios,the processing of the predicted results improves the accuracy.Training and testing in a distributed manner.