Label filtering method, system, device and storage medium for hotel risk pictures
Through the multimodal recognition method, combined with the Q&A tree and neural network model, the risk facilities and text in the hotel risk pictures are identified, and the problem of incomplete and inaccurate identification in the existing technology is solved, and efficient and reliable risk picture recognition and filtering is achieved.
CN119760100APending Publication Date: 2025-04-04CTRIP TRAVEL NETWORK TECH SHANGHAI0
0 Cites 0 Cited by
Patent Information
- Application Number
- CN202411975109.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-04
AI Technical Summary
Technical Problem
The prior art is difficult to effectively identify and classify hotel risk images containing both image and text information, resulting in incomplete and inaccurate identification.
Method used
Using multimodal recognition method, by establishing a recognition system based on a question-and-answer tree and training neural network model, end-to-end identification of risk facilities and text, combined with image editors and large language models, efficient filtering of massive hotel photos is achieved.
Benefits of technology
It improves the degree of automation of risk image recognition, reduces dependence on manual audits, reduces operational costs, and ensures the comprehensiveness and reliability of identification results.
✦ Generated by Eureka AI based on patent content.
Abstract
The present invention provides a label filtering method, system, device and storage medium for hotel risk pictures, the method comprising: collecting hotel pictures; establishing a question-answering tree based on which the hotel pictures are first identified, each level of the question-answering tree comprising a preset risk question, at least one preset risk threshold and at least one preset risk label, after the hotel pictures are identified based on the preset risk question, if the feedback answer meets the preset risk threshold, the preset risk label corresponding to the preset risk threshold is added to the label set corresponding to the hotel picture; a neural network model is trained based on the hotel pictures with the label set; a large number of hotel photos are input into the trained neural network model for filtering. The present invention can simultaneously identify risk facilities and text in an end-to-end manner, while improving the recognition efficiency, ensuring the comprehensiveness and reliability of the recognition results.
Need to check novelty before this filing date? Find Prior Art