Container number localization based on SSD algorithm

After R-CNN and other classical frameworks were proposed, deep learning based target positioning framework has gradually become the mainstream. The SSD (Single Shot MultiBox Detector) framework model was used to locate the container number. By comparing the locating effects of different labeling methods for training samples, the basic unit that label as Single row or Single column is much better. With the actual size and proportion of the container number as reference, that modifying the feature layers which are used in SSD frame to generate the default box and adjusting the size of the feature map could adapt to the localization better. The result show that in the complex container number scenario, the accuracy based on SSD network to detect container number reaches 78%.