Domain Adaptive Ship Detection in Optical Remote Sensing Images

With the successful application of convolutional neural network (CNN), significant progress has been made by CNN-based ship detection methods. However, they often face considerable difficulties when applied to a new domain where imaging condition changes significantly. Although training with two domains together can solve this problem some extent, large shift will lead sub-optimal feature representations, and thus weaken generalization ability on both domains. In paper, adaptive method is proposed better detect ships between different Specifically, minimizes discrepancies via image-level adaption instance-level adaption. adaption, we use multiple receptive field integration channel attention enhance feature’s resistance scale environmental changes, respectively. Moreover, novel boundary regression module in correct localization deviation proposals caused shift. Compared conventional approaches, able make more accurate predictions effective extreme point features. The components are implemented learning corresponding classifiers respectively an adversarial way, thereby obtaining robust model suitable for Experiments supervised unsupervised scenarios conducted verify effectiveness method.