Research And Implementation Of Secret Picture Recognition Algorithm Based On Deep Learning

With the development of science and technology,e-government is gradually replacing the traditional way of handling government affairs.While bringing convenience to people,it also brings the risk of leakage.Many leakage risks,such as office computer poisoning and illegal copying of confidential data,threaten our computer system security and state secrets.The TIPTOP document operation security audit system is a system that specifically monitors and filters confidential information on government office computers.Due to technical limitations,this system can only perform confidential identification for text content and cannot identify image content.Today,more and more information has been stored in picture and video.Due to technical limitations.Therefore,it is imperative to develop an algorithm for identifying secret pictures.So this paper uses deep learning technology to design secret image recognition algorithms and improve the document auditing system.The secret image recognition algorithm performs scanning and confidentiality determination on the input picture,outputs the secret picture,and marks the location of the secret information in the picture.The main work of this article includes:1?Introduce TIPTOP document operation security auditing system,combined with this system can not identify the limitations of the picture,put forward the idea of implementing the secret image recognition algorithm.2?Describe the principles of deep learning and convolutional neural networks.By comparing the large differences in accuracy between convolutional neural networks and traditional image recognition algorithms in ILSVRC,the correctness of using deep learning is demonstrated.Research the object recognition algorithm based on deep learning,and introduce the concept and internal structure of common target recognition algorithms such as Faster R-CNN,R-FCN,YOLO,and SSD.Then,the advantages and disadvantages of these algorithms are introduced through the comparison of different algorithms in mAP,training time,different picture sizes andresolutions.Finally,by comparing the results,Faster R-CNN and SSD were selected as the final model.In addition,the influence of different hardware on training time was also studied,and the advantages of using GPU for calculation were clarified.3? Based on the principle of transfer learning,the algorithm of secret image recognition is completed through selection of experimental software and hardware,collection and processing of data sets,selection of pre-training models and model training.Finally,the advantages of Faster R-CNN relative to SSD in secret image recognition algorithms are analyzed,and the impact of implementation of secret image recognition algorithms on document auditing system is summarized.