In the context of digital transformation of libraries, traditional manual management models have problems such as low efficiency and large errors. Existing intelligent recognition technologies face bottlenecks such as insufficient collaborative optimization between text recognition and object detection, accumulated positioning errors, and difficulty balancing real-time performance and accuracy. In response to the above issues, this study proposes a robotic arm control method based on the optimized "You Only Look Once" algorithm 11th edition (YOLO11) and Convolutional Recurrent Neural Network (CRNN). A global context enhancement module is integrated in YOLO11 to enhance target perception in complex backgrounds; improves the CRNN architecture, adds attention mechanism to achieve dynamic focusing of text regions, and designs a character correction process in conjunction with the Chinese Library Classification to effectively solve the problem of misjudgment of similar characters. The experimental results show that the improved model achieves a book spine recognition accuracy of 99.3% at a resolution of 1024 × 1024, which is 1.9 percentage points higher than the original YOLO11; the average accuracy under different lighting conditions (94.2% - 97.8%) is significantly better than comparative algorithms; the joint motion parameters of the robotic arm have small fluctuations. In practical applications, the average classification accuracy of six types of books is ≥ 91%, and the average shelf time is about 240.7 seconds, meeting the efficiency and accuracy requirements of library automation management. Therefore, the proposed multimodal fusion method and closed-loop control architecture significantly improve the reliability and positioning accuracy of library automation systems, providing an effective solution for intelligent book management in complex scenarios.
LI et al. (Thu,) studied this question.