Our investigation aims at training deep networks for monitoring waterway traffic means on the rivers in the Mekong Delta. We collected the real videos of the waterway traffic, and then tagging the five most popular means in frames extracted from the videos, making an image dataset. We propose to train recent deep network models such as YOLO v4 (You only look once), RetinaNet and EfficientDet on this image dataset to detect the five most popular means in the videos. The numerical test results show that YOLO v4 gives highest accuracy than two other methods, including RetinaNet and EfficientDet. YOLO v4 achieves the performances on the testset with a precision of 91%, a recall of 98%, F1-score of 94% and mean average precision (mAP@0.50) of 97.51%.