Research And Implementation Of Dishes Identification Technology In Unattended Settlement System

In recent years,the concept of “new retail” has set off a new wave of development in all facets of the sales industry.As one of the traditional sales service industries,although the catering industry maintains a good development speed,it still faces some difficulties and challenges: high operating costs,low profits,and high update speed.Nowadays,the “new retail” of catering is mainly represented by catering informatization and automation.In terms of payment and settlement,the unattended settlement system is a major achievement of catering automation,which effectively reduces the labor cost and operational efficiency of the restaurant.In order to build a restaurant unattended settlement system,the first step is to realize the identification of dishes.Fortunately,the development of deep learning technology provides new technologies and reliable solutions for dish identification tasks in unattended settlement systems.However,deep neural network based identification systems typically use a single network to implement dish detection and classification.The network is often a fixed structure.Therefore,when adding or subtracting identification dishes varieties,we not only need to modify the output types in the network structure,but also need to collect a large number of dishes to retrain the network,which greatly increased the cost of update.In view of the shortcomings of this traditional dish identification scheme,this paper proposes a dishes identification technology based on deep learning.The proposed technology includes two parts: object detection and image retrieval.Firstly,we determine the location of the dishes in the image through a dish detection network,and the dishes are separated from the image background according to the dish locations.Then,the feature extraction network is used to calculate the image features of the separated dish images.We search in the reference library of dishes to obtain the dish classification results based on the image feature.The reference library of dishes used in this paper is an image feature set of pre-registered dishes.Finally,the identification system output the location and category information of dishes.This paper uses different data sets according to the requirements of object detection network,image feature extraction network and reference library of dishes.Since the feature extraction network needs high generalization ability,this paper selects training samples from Chinese foodnet dataset,which contains 208 kinds of scene-rich dish images.In addition,we build the training data set of object detection and the image data of reference dish library.The former contains 528 dish images with background,and the latter contains 105 classes of separated dish images.The proposed dish identification technology in the unattended settlement system only needs to update the reference menu to realize the increase and decrease of the system identifiable dish variety.It does not need to retrain the feature extraction network,which solves the problem that the current algorithm of dish identification is not easy to update.This technology reduces the cost of updating the dish identification algorithm,which makes it easier to be applied and popularized,and is more in line with the needs of “new retail” catering industry.