Design of a Text Recognition System Based on Jetson Nano

In laboratory environments with high confidentiality requirements, traditional manual data recording methods are inefficient and prone to errors. To address this challenge, this paper proposes an offline text recognition platform based on the Jetson Nano. The platform uses embedded devices for text recognition, supporting batch image uploads from terminal devices and providing fast recognition results. The core algorithm uses PaddleOCR, and the model is optimized and pruned through Paddle Lite to fit the computational and memory limitations of the Jetson Nano. Additionally, the platform leverages GPU acceleration to perform parallel image preprocessing, significantly improving processing efficiency. Experimental results show that, when applied to the Jiangmen neutrino experiment, the platform processes on-site instrument data images at a speed of 1 seconds per image with a recognition accuracy of ${9 4. 5 \%}$, and supports simultaneous upload requests from four terminal devices. The platform not only improves data collection efficiency but also ensures data security, making it suitable for text recognition needs in high-security environments.