Automatic container code recognition (ACCR) plays an important role in customs logistics and transport management. Due to complicated lighting conditions and background pollution, automatic detection and recognition of container codes remains a difficult task. In this work, we exploit Faster-RCNN, a robust object detection algorithm based on deep learning algorithm, to detect and recognize container codes. First, container code characters are detected as 36 classes of small objects, consisting of 26 capitals and 10 digits. Next, a novel post processing algorithm based on binary search tree is adopted to find container code from detected characters. Experimental results validate the proposed approach, and the overall accuracy on a dataset with 831 container codes achieves 97.71%.