Flag detection method, system and equipment based on high-confidence labeling strategy

Abstract

The invention discloses a flag detection method, system and equipment based on a high-confidence labeling strategy. Flags in pictures are detected through core feature information of different target flags. Firstly, related flag pictures are collected; 2, expanding unbalanced flag class pictures; thirdly, constructing a core feature labeling standard; fourthly, constructing a complete data set; 5, training a supervised flag detection model by using the constructed data set; and finally, identifying the flag in the unknown picture by the detection model. According to the method, the core feature information of the target flag is labeled by using the core feature labeling criterion, so that the core feature information of the target flag is better obtained, the confidence coefficient of the labeled sample is improved, the detection capability of the model on different flag picture categories is improved, the problem of recognition of shielded flags and deformed flags in the pictures is better solved, and the recognition efficiency is improved. And meanwhile, an effective data enhancement method is used for processing the problem of data category imbalance, and the method has the advantages of high recall, high robustness, high efficiency and the like.

Description

technical field [0001] The invention relates to the field of target detection, in particular to a flag detection method, system and equipment based on a high-confidence labeling strategy. Background technique [0002] Flag detection refers to the technology of detecting the location and category of the target flag from a picture. Flag detection technology is widely used in the field of image content review. On major Internet platforms, users upload a large number of pictures every day, which may contain some sensitive flags. Flag detection belongs to the field of target detection, and its supervised learning characteristics lead to high requirements for original training data, and the labeling work is the early basis of target detection, and the quality of labeling will directly affect the effect of target detection. At present, the data labeling criterion in the target detection task is mainly to globally label the target object, that is, to use a rectangular frame to labe...

AI Technical Summary

Problems solved by technology

[0004] The above-mentioned flag detection method still uses the traditional global labeling method to mark the target flag, without considering the core features of different flags, and cannot fully extract useful discriminative information. At the same time, the global labeling method will introduce a lot of noise information, making the labeled samples The confidence is low, especially in scenes such as flag occlusion and deformation, which makes it difficult to detect the target flag

Images