Deep neural network model marking and recognition method and system based on blind watermarks

Abstract

The invention provides a deep neural network model marking and recognition method and system based on blind watermarks, and the method comprises the steps: taking a basic image and an exclusive mark as the input, generating blind watermarks with invisible features, and enabling the distribution of the blind watermarks to be completely consistent with the distribution of the basic image and not tobe differentiated; assigning a predefined label to the generated blind watermark, then mixing the blind watermark and the training data set into a batch of samples according to a certain proportion, sending the batch of samples into a set deep neural network model for training, enabling the blind watermark to be embedded into the deep neural network model to form a host model, and realizing modelmarking; according to the invention, the set mark can be embedded into the normal image to generate the blind watermark with invisible features, and then the blind watermark is embedded into the hostmodel to be protected, so that marking and recognition of the network model are realized.

Description

Technical field [0001] The present disclosure belongs to the technical field of information marking and identification, and relates to a deep neural network model marking, identification method and system based on blind watermarking. Background technique [0002] The statements in this section merely provide background information related to the present disclosure, and do not necessarily constitute prior art. [0003] Deep learning technology has made great progress in various challenging tasks such as computer vision, natural language processing and autonomous driving. As the core part of deep learning technology, deep neural network plays a vital role in its development. Most major technology companies use deep neural network as a key component to build their artificial intelligence products and services. [0004] Building a high-performance deep neural network model is still a very arduous task, requiring more time and wisdom from technicians. However, models and images are diffe...

AI Technical Summary

Problems solved by technology

[0004] Building a high-performance deep neural network model is still a very difficult task, requiring more time and wisdom from technicians, but the model is different from the image and is not intuitive. When identifying whether others have used the depth of their own design / build Neural network model, there is a greater difficulty

Images