Ship detection technology has always played an important role in operations such as maritime traffic management, ship protection and identification of noncompliant ships. When conducting water surveillance, ship target detection is carried out not only to ensure ship safety but also to supervise non-compliant ships. Due to the current improvements in physical imaging technology and developments in the field of machine vision, it is now possible to use deep learning methods to train neural networks for ship inspection. In all types of ship inspection tasks, ship targets are variable in direction and vary greatly in scale, and the detection accuracy is not high if the generic traditional horizontal frame detection method is made to work. Therefore, this paper proposes the HSVAugRegNet based on the framework of RegNet, which deepens the network depth based on RegNet and also adds data enhancement means HSV random perturbation to change the image brightness and improve the network generalization ability. After conducting experiments on the well-known dataset HRSC2016, it was found that compared to the original method, the results obtained by HSVAugRegNet improved 3.41% of the map, satisfying the need to improve recognition accuracy.
HSVAugRegNet: Random Augmentation to Rotate Image Recognition
28.04.2023
1950276 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
British Library Conference Proceedings | 1999
|A Quiet Sunday - Slow to rotate
Online Contents | 2006