Fast and Lightweight Automatic Shipping Container Attributes Spotting

The surge in global trade and container port throughput underscores the critical need for precise and efficient management of shipping containers. Current automated approaches primarily focus on container codes, overlooking essential attributes, such as types, weights, and hazard signs, and often fail to accommodate texts in various forms, including horizontal, vertical, and multi-line. In this article, a fast and lightweight automatic shipping container attributes spotting (FLASS) is proposed for the first time to detect and recognize arbitrary-oriented container codes, types, weights, and hazard signs. This framework is based on three optimized deep-learning models: YOLOv8 Super Light (YOLOv8-SL) for object detection, Super Light Text Detection (SL-TD) for text detection, and Robust SVTR (SVTR-R) for scene text recognition. Extensive experimentation demonstrated that FLASS, comprising these three distinct models, achieves remarkable accuracy of 96.29%, offers high-speed inference at 32.71 FPS, and utilizes a lightweight model with just 10.3 million parameters. The core concept of our method has been deployed and operational at gates, tallying stations, and gantry cranes at Ningbo Zhoushan Port for months, demonstrating practical application capabilities.