The School of Civil Engineering, Harbin University, Harbin 150086, China.
Comput Intell Neurosci. 2022 Sep 21;2022:4854213. doi: 10.1155/2022/4854213. eCollection 2022.
The current work aims to meet the needs of the development of archives work in colleges and universities and the modernization of management to realize the standards and standardization of all aspects of archives business construction in colleges and universities, so as to improve the political and professional quality of archives cadres. First, the radio frequency identification (RFID) technology based on the Internet of things (IoT) digitizes the university archive labels. Meanwhile, the filing cabinet's intelligent security system preserves confidential files. Second, the convolutional neural network (CNN) algorithm under deep learning is introduced and college profile information is identified. Finally, the concept of professional certification is used to clarify the purpose of the university archives automation management system. Different activation functions are used to analyze the recognition accuracy loss and recognition accuracy of university archives. The identification error of You Only Look Once (YOLO) of the ReLU-convolutional neural network (R-CNN) of college archives is analyzed. The results show that the selection of rectified linear units (ReLU) activation function for CNN can effectively reduce the loss of identification accuracy of college archives and can improve the accuracy of identification of college archives. The algorithm based on the ReLU activation function has a smaller recognition error accuracy in college archives than that of the YOLO algorithm. The recognition error of the YOLO algorithm is slightly higher than that of the R-CNN. The font recognition error of archival information based on the R-CNN is relatively large. However, the conclusion is reasonable due to the recognition difficulties of handwritten archival fonts. The file positioning recognition error rate is 19.00%, the file printing font recognition error rate is 4.75%, and the image recognition error rate is 1.90%. These results have a certain reference value for the process of identifying information in the automatic management of university archives by CNN under different activation functions.
当前的工作旨在满足高校档案工作发展和管理现代化的需要，实现高校档案业务建设的各项标准和规范化，提高档案干部的政治和专业素质。首先，基于物联网 (IoT) 的射频识别 (RFID) 技术对高校档案标签进行数字化。同时，文件柜的智能安全系统保存机密文件。其次，引入深度学习下的卷积神经网络 (CNN) 算法，识别高校档案信息。最后，利用专业认证的概念，明确高校档案自动化管理系统的目的。使用不同的激活函数来分析高校档案的识别精度损失和识别精度。分析了 YOLO 算法在 ReLU 卷积神经网络 (R-CNN) 中的识别误差。结果表明，CNN 中选择修正线性单元 (ReLU) 激活函数可以有效地降低高校档案识别精度的损失，提高高校档案的识别精度。基于 ReLU 激活函数的算法在高校档案中的识别误差精度比 YOLO 算法小。YOLO 算法的识别误差略高于 R-CNN。基于 R-CNN 的档案信息的字体识别误差较大。然而，由于手写档案字体的识别难度较大，因此该结论是合理的。文件定位识别错误率为 19.00%，文件打印字体识别错误率为 4.75%，图像识别错误率为 1.90%。这些结果对于不同激活函数下 CNN 识别高校档案自动管理过程中的信息具有一定的参考价值。
