Design And Implementation Of Traceable Mini Program Based On Deep Learning

With the increasing proportion of digital products in GDP,the illegal spread of digital products has brought serious economic losses to their owners.In order to strengthen the protection of digital products,researchers have developed many applications for traceability and leakage prevention,the main principle of which is based on information hiding algorithms.The traditional information hiding method is only effective in the field of its design,and basically does not have the generalization the protection of digital products.The study found that the neural network is highly sensitive to small disturbances in the input image,and this function can be used to hide information.In order to solve the problem of image distortion cuased by rotation and perspective transformation in the transmission process,this paper designs a trace source mini program system based on the deep learning model.The system includes an encoder,a detector and a decoder.The encoder is responsible for converting the incoming pictures and the watermark information that needs to be hidden into encoded pictures(pictures containing watermark information and anchor points),and the detector is based on the designed lightweight object detection network and is responsible for detecting from the uploaded encoded pictures Anchor point.The decoder mainly performs geometric correction on the image based on the anchor point and the picture,and then uses the decoder network to obtain the watermark information from the corrected picture.In order to enhance the accuracy and practicability of detection,based on the idea of separation of front and back ends,using the design idea of We Chat applet that is ready to use and go,the detector is designed into a convenient management applet system,through which the upload By comparing the decoded ciphertext with the database,the source information hidden in the picture is obtained to achieve the purpose of tracing the source and preventing leakage.In the testing phase,the m AP@0.5 of the detector model on the self-made VOC test set is more than 90,and the Recall can be more than 95.In the actual test,the mobile phone is used to take pictures,and then the applet system is used for detection.The results show that in most cases,the system can accurately identify the user information hidden on the screen.