Research on Intelligent Recognition Algorithm of Container Numbers in Ports Based on Deep Learning

The identification of container number has important application value in the field of logistics and cargo transportation. A new container number recognition algorithm was proposed in this paper to solve the difficult problems such as different illumination conditions, blurred image, loud noise, damaged and polluted container number, zigzag deformation, etc. First, the low-light enhancement algorithm based on Retinex theory was used to process the container number image to deal with the problems of inconsistent port lighting conditions and background noise. The super-resolution reconstruction was used to deal with the problems of container surface contamination and container number damage. The backbone network was replaced by MobileNetv3 by improving the YOLOv5 algorithm. The ECA attention mechanism was added to achieve lightweight model and accurate location of box number area. STN is added before the convolutional layer of the CRNN to correct the image. Public images on Github and official images of Tianjin Port were used to generate samples through DCGAN network, and their data were enhanced. The obtained 6961 container number images were used as data sets to train the improved CRNN model. The mAP of the proposed method in container number location using the improved YOLOv5 reaches 93.7%, the accuracy rate reaches 94.5% in container number identification using the improved CRNN, and the average recognition speed reaches 29.1 frames/s. The method performs well in real-time performance and realizes the lightweight of the model. It can meet the requirements of port real-time and accurate identification of container number.