Residual YOLOX-based Ship Object Detection Method
Ming Liu, Changming Zhu
Abstract
With the development of the times, Shipping and logistics-related industries have higher requirements for the accuracy and speed of ship detection algorithms. This paper proposes a ship object detection method based on YOLOX algorithm, which can be used for ship image detection in ports. Refer to the COCO dataset format, construct the ship object detection data set, introduce the residual structure and the CIOU loss function to improve and optimize the algorithm, and compare the model performance with the original YOLOX algorithm. The test results show that the Res-YOLOX algorithm is significantly better than the original YOLOX algorithm, for a single class ship object detection, the AP <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5</inf> on the test set reached 88.7%, increased by3.5%, and for multiclass ship object detection, the AP <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5</inf> reached 83.6%, 54.3%, 40.3% respectively. Finally, model compression is used for trained model. After compressing model, the speed is increased by 15.4ms.