Library archive management is an important business in the library management system, and the progress of science and technology has made the library archive management system increasingly intelligent. The embedded system has strong real-time and specificity, thus making it suitable for the development of archive management systems. The article introduced the composition and structure of the embedded system, including software and hardware parts. To further improve the efficiency of embedded archive management systems, the article used a deep learning convolutional neural network algorithm model to improve it. The performance of the convolutional neural network algorithm was tested from three aspects: the accuracy of archive system risk prediction, the system’s transaction processing ability, and the system’s clicks per second. The data showed that the risk prediction accuracy of embedded archive systems under convolutional neural networks was above 0.9. At the 80th experiment, the risk prediction of traditional embedded archive systems decreased to 0.89. The convolutional neural network embedded archive system processed more transactions within a certain response time. The convolutional neural network embedded archive system had multiple clicks during a certain period of time, with a maximum of about 60 clicks per second. The overall click through rate of traditional embedded file systems was below 50 times per second. Therefore, it could be concluded that embedded archive systems based on convolutional neural networks had better transaction processing capabilities and work efficiency.
