Cigarette identification method suitable for wide retail scene

Abstract

The invention discloses a cigarette identification method suitable for a wide retail scene. The method comprises the following two steps of A, carrying out image sample collection on various kinds ofexisting cigarettes in a fixed scene, and finishing fusion of actual retail scenes and data enhancement for cigarette sample images through an automatic scene fusion algorithm; and B, by utilizing data of the fused actual scenes, in combination with cigarette characteristics, carrying out identification model training to obtain a cigarette identification model. According to the method, by fusing actual scene images, the actual scene data in cigarette detection and identification can be rapidly and conveniently enhanced, so that the scale and the scene diversity of training data are improved; the method can be used for training collected data in a new scene; and the method is low in cost, is simple, convenient and rapid, is high in adaptability, high in identification precision and obviousin effect, and can save a lot of labor and financial cost expenditures.

Description

technical field [0001] The invention relates to an image recognition technology, in particular to a cigarette recognition method suitable for a wide range of retail scenarios. Background technique [0002] Image recognition is an important field of artificial intelligence. To put it simply, image recognition is a technology that uses computers to process, analyze and understand images to identify targets and objects in various patterns. In recent years, with the development and popularization of artificial intelligence, especially the extensive research and development of deep learning technology, image recognition technology has also entered a new stage. The main method of image recognition is from the previous feature extraction joint statistical learning. The two-step method gradually shifts to an integrated method of autonomous extraction and learning of deep features based on convolutional feature extraction. [0003] In the process of selling cigarettes, major tobacc...

AI Technical Summary

Problems solved by technology

The current statistics of cigarette retail data are mainly in the sales process, and the salesperson uses the code scanning method to complete the statistics of cigarette data. This statistical method has the disadvantages of cumbersome operation and easy omission of scanning, and it is difficult to establish the relationship between cigarettes and customers who buy cigarettes. Through image recognition, it can not only automatically identify cigarettes and customers in the sales process, thereby simplifying the workload of sales staff, avoiding the risk of missing scans, but also effectively establishing the data relationship between cigarettes and buyers , to provide a reliable data source and basis for subsequent data analysis

[0004] In the actual cigarette image recognition, there will be two problems that mainly affect the accuracy of cigarette recognition: First, the environment and scenes of each retail point are different. Under the premise of the same recognition model structure, the scale of training data and the diversity of scenes , will have a greater impact on the accuracy and performance of the recognition

Second, the appearance of some cigarettes is basically the same, and there are only significant differences in some areas. Such similar cigarettes have an impact on the accuracy of the overall image recognition

Images