Ship target detection and identification based on SSD_MobilenetV2

There are many deep learning algorithms currently used in ship supervision, but they generally have the problems of insufficient target detection speed and accurate identification rate. In this paper, an improved SSD algorithm based on MobilenetV2 convolutional neural network is proposed for ship image target detection and identification. MobilenetV2 network is used for feature extraction of ship images. The MobilenetV2 network is pre-trained on the Coco dataset, and then fine-tuned on the constructed ship image dataset to save training time and computing resources. In order to verify the accuracy of the SSD _MobilenetV2 algorithm in ship image detection and id0entification, Faster R-CNN_InceptionV2 algorithm was used as a comparison group. Experimental results show that the SSD_MobilenetV2 algorithm has better detection and identification effect on ship images. And the SSD_MobilenetV2 algorithm has better performance in terms of detection speed and NMS(Non-Maximum Suppression), which meets the requirements for rapid detection and identification of ship targets.