Catering kitchen violation judgment method based on Encoder-Decoder model and Gaussian mixture model

Abstract

The invention discloses a catering kitchen violation judgment method based on an Encoder-Decoder model and a Gaussian mixture model, and relates to the technical field of image processing. The methodcomprises a modeling stage: preparing violation pictures, and constructing a violation picture set; marking violation regions in the violation picture, and constructing a violation region set; constructing an Encoder-Decoder model, wherein the Encoder-Decoder model comprises an encoder Encoder and a decoder Decoder; training an Encoder-Decoder model by taking the violation graph set as a sample until the convergence of a cost function reaches the standard; and constructing a Gaussian mixture model by taking the coding result of the trained encoder Encoder as input, and taking the output resultas the violation probability. An implementation stage: extracting an encoding result of the picture to be detected by using the trained encoder; inputting a coding result into the Gaussian mixture model, and calculating the violation probability; and judging whether a violation behavior exists in the to-be-detected picture or not according to the violation probability. According to the method, violation visual features can be accurately extracted. The method is not limited by kitchen scenes, and violation identification generalization performance for different catering stores is good.

Description

technical field [0001] The invention relates to the technical field of image processing, and relates to a method for judging irregularities in a restaurant kitchen based on an Encoder-Decoder model and a mixed Gaussian model. Background technique [0002] Computer vision is a technology that has matured in recent years. Intelligent video surveillance technology based on computer vision technology is widely used in restaurants, companies, gyms, construction sites, and railway stations. With the improvement of people's living standards, food and beverage The requirements of the industry have also increased. In order to make the management of the kitchen in the catering industry more and more standardized, the catering intelligent video monitoring technology based on computer vision technology has also emerged as the times require. [0003] The general practice of catering intelligent video surveillance based on computer vision technology is to firstly use the method of target ...

AI Technical Summary

Problems solved by technology

[0004] However, the current algorithm has certain limitations. First of all, in the back kitchen scene of the catering industry, the visual background is messy and there is a large amount of redundant interference information, which will cause certain interference to the target detection algorithm and cause the performance improvement of the target detection algorithm to be affected. Secondly, based on the target detection algorithm to detect the target combined with some semantic information of the kitchen, artificially formulate corresponding rules to find out the violation items in the kitchen inspection standard. When the processing rules are simple and the visual features are single and clear, the intelligent algorithm can Accurately find problems, reduce the number of video views, and improve efficiency

However, many inspection items in the back kitchen cannot be formulated with clear rules. As a result, the intelligent algorithm can only find some suspected violations, and the false positive rate will be relatively high, which is of limited help to improve work efficiency.

Images