<h1 class="pgc-h-arrow-right" data-track="1" > This article is from the WeChat public account: X-MOLNews</h1>
英文原题：Paper Information Recording and Security Protection Using Invisible Ink and Artificial Intelligence
Corresponding Author: Li Kang, Harbin Institute of Technology (Shenzhen); Xu Jie, Harbin Institute of Technology; Zhao Weiwei, Harbin Institute of Technology
作者：Yunhuan Yuan, Jian Shao, Mao Zhong, Haoran Wang, Chen Zhang, Jun Wei，Kang Li, Jie Xu and Weiwei Zhao
We live in an information age in which information can be disseminated in a variety of ways, including television, radio, the Internet, and print materials. Despite the growing importance of electronic media, paper still plays a vital role in the dissemination and storage of information. However, with the development of the times, paper information faces serious security challenges. The leakage of information can lead to serious economic and social problems. In view of this, it is necessary to develop an effective strategy for encrypting/decrypting the information on the paper. Currently, the most common method is to use ink with unique optical properties to write or print information on paper. This information is only displayed during external stimuli such as radiation, chemical treatment, or heating. However, this method, which relies solely on the encryption of the properties of the material itself, is less complex and predictable. Once the nature of the ink is exposed, the information can be easily cracked, posing a significant risk to commercial and military applications. While fluorescence microplate readers have been used to enhance the security of information, the complexity and unpredictability of encryption still need to be improved. Artificial Intelligence (AI) has produced a wide range of applications in areas such as medicine, mobile health surveillance, military decision-making, and cybersecurity, bringing huge changes to people's lives. In recent years, it has also been gradually introduced into materials chemistry, such as for the synthesis of organic molecules, identification of thermodynamic stability of 2D materials, etc. However, to the best of our knowledge, there is currently no work combining nanomaterials and artificial intelligence to encrypt and decrypt paper information.
In order to improve the security and unpredictability of paper information, Professor Weiwei Zhao's team at Harbin Institute of Technology proposed a strategy for combining Carbon nanoparticles (CNPS) and artificial intelligence for high-level information security protection (Figure 1). In this work, a fluorescent quantum yield (QY) of high quantum yield (QY) was prepared by microwave method, and a carbon nanoparticle with high stability and low cytotoxicity. Subsequently, the nanoparticles are prepared as invisible inks for secure encryption of paper information. To improve the complexity and unpredictability of information, researchers have built a five-layer convolutional neural network (one of two mainstream models in today's field of artificial intelligence) as an auxiliary tool to encrypt information at a deeper level. The correct information can only be obtained by irradiation with ultraviolet rays and input to a specially trained neural network, otherwise it will output wrong and misleading content. Since the neural network is a black box, the training information is hidden in its millions of parameters, and it is almost impossible to find the corresponding cipher book from the network structure itself alone. As a result, AI provides extremely secure protection for paper information.
Figure 1. Schematic diagram of the entire process of obtaining, encrypting, and decrypting paper information.
In the first step of the work, the researchers prepared fluorescent carbon nanoparticles with high quantum yield using microwave method, and then basically characterized the basic properties of the material (Figure 2), including basic morphology, ultraviolet absorption test, fluorescence excitation and emission spectra, crystal structure, surface functional groups, etc.
Figure 2. Characterization of carbon nanoparticles.
Then, to enable applications in complex environments, the researchers examined the stability of carbon nanoparticles (Figure 3). As can be seen from Figure 3, prolonged UV lamp exposure has little effect on the fluorescence intensity of the material. At the same time, the fluorescence intensity of the material is virtually unchanged, whether after a long period of placement (140 days) or exposure to a medium of high ionic intensity. Finally, the nanoparticles are virtually non-toxic to cells. These results show that in different environments, the optical properties of CNPS are not affected, and it has strong stability, which is very beneficial for its application in the fields of imaging, biological sensing and security.
Figure 3. Stability and toxicity of carbon nanoparticles.
Next, the researchers made carbon nanoparticles into invisible ink and used artificial intelligence as an auxiliary tool for the recording and security protection of paper information (Figure 4). The training samples were obtained by inkjet printing fluorescent ink and UV lamp irradiation (Figures 4a and 4b). These samples and their corresponding labels are then used to train and validate the AI model (Figure 4c). The test set is also obtained by inkjet printing and UV lamp irradiation, the only difference being that both ordinary ink and invisible ink are used to complete the printing (Figures 4d and 4e).
Figure 4. AI model training and test samples.
Convolutional neural networks (CNNs) are one of the two mainstream models in the field of artificial intelligence, which have been used in various fields, including speech recognition, natural language processing, computer vision, etc., and have easy-to-use tools to build and train. This article uses CNN for the encryption and decryption of paper information, and its basic structure is shown in Figure 5a. The model training process is completed on a personal computer. Usually after 300 iterations of training, the network can learn the structural information of all symbols and achieve nearly 100% accuracy on both the training and validation sets (Figures 5b and 5c). The test sample can be displayed as a set of symbols in natural light and can also be fed normally into the network to produce an output (Figure 4d); however, the output information is incorrect (STOP) due to incomplete symbols displayed in natural light. The correct symbol can only be obtained under ultraviolet radiation (Figure 4e), and entering it into the network will produce the correct information (BEGIN).
Figure 5. Structure and accuracy of AI models.
In summary, the correct information can only be extracted after UV exposure and the information is fed into a specific trained neural network. This approach is more complex and unpredictable than traditional paper-based information protection methods that rely primarily on stimulus-responsive materials, and is therefore more difficult to crack. The encryption and decryption methods proposed by the Institute can meet all the requirements of paper information protection, including easy preparation, low cost, high stability and high security.
The paper was published in ACS Applied Materials & Interfaces, with Yunhuan Yuan, a doctoral student at Harbin Institute of Technology, as the first author, and Professor Weiwei Zhao, Professor Xu Jie, and Assistant Researcher Li Kang as co-corresponding authors.
Paper Information Recording and Security Protection Using Invisible Ink and Artificial Intelligence
Yunhuan Yuan, Jian Shao, Mao Zhong, Haoran Wang, Chen Zhang, Jun Wei, Kang Li*, Jie Xu*, and Weiwei Zhao*
ACS Appl. Mater. Interfaces, 2021, DOI: 10.1021/acsami.1c01179
Publication Date: April 20, 2021
Copyright © 2021 American Chemical Society
Instructor introduction
Zhao Weiwei
https://www.x-mol.com/university/faculty/302040
(This article is from ACS Publications)