Image detection method and device

Abstract

The invention discloses an image detection method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the steps of extracting an embedded featureof a first target image; determining category information corresponding to the embedded features of the first target image through a normalization layer and an image classifier; and according to thecategory information determined by the normalization layer and the category information determined by the image classifier, outputting a detection result of whether the first target image is a specific category image. The detection model is high in training speed, has the capacity of rapid iterative optimization, can rapidly respond to online image false detection conditions, is high in detectionefficiency, strong in generalization capacity and high in mobility, can achieve a very good detection effect in a new service scene, and fully meets the detection requirements of multiple service scenes.

Description

Technical field [0001] The invention relates to the field of computer technology, in particular to an image detection method and device. Background technique [0002] At present, many websites and apps (applications) provide the function of UGC (user-produced and published content), so that each user can publish self-produced and edited multimedia information, and the difficulty of monitoring online multimedia content by platforms has increased sharply , The existing manual review methods are facing huge challenges. At the same time, the multimedia content has various forms, including text, images, audio, video, etc., which also puts forward higher requirements for reviewers. [0003] Existing content review mainly includes two schemes: one is to use image classification convolutional network as the detection model, and the core structure of Inception (GoogLeNet (a deep network structure) is used to increase the depth and width of the network). , Resnet (residual error network) tw...

AI Technical Summary

Problems solved by technology

[0006] The first scheme takes a long time to train the detection model. When there are false detections (such as missing or killing) violations online, the speed of the iterative optimization model is slow, so that it cannot respond quickly after the false detection image appears, which affects the detection efficiency. , although the second scheme can achieve rapid iterative optimization by expanding the illegal image database, it does not have the generalization ability and cannot intercept images other than the image database, and the above two schemes are not highly portable, and it is difficult to meet multi-service Scenario detection requirements

Images