Research On Maritime Target Fusion Detection In Multi-source Remote Sensing Images Based On R-YOLO

Detection of sea surface targets in large-scale remote sensing images is one of the important research topics of ocean remote sensing technology.Rapid and accurate detection of ship targets from remote sensing images is an important research direction in the field of target detection.However,remote sensing images have the characteristics of wide format,strong interference and small target.This thesis mainly focuses on how to effectively improve the anti-interference ability of remote sensing image ship detection algorithm,the accuracy and recall rate.To solve the problem of high false alarm probability and low recognition rate in traditional large-format complex scene remote sensing image ship detection algorithm,this thesis applies the deep learning target detection network YOLOv3 which learns and predicts the location information of ships in different backgrounds through training in remote sensing ship dataset.To improve the recall rate of the target detection algorithm,this thesis adopts the spinning target detection method,and proposes a ship detection model based on R-YOLO.Through redefining the representation of the rotation matrix and redesigning a new network loss function and the rotated IOU computing method,this model accurately outputs the real length,width and axial information,increases the output feature dimensions,and effectively raises the recall rate and speed of multi-target detection.Lastly,to improve the practicability of the algorithm on mobile devices,the model is processed in a lightweight way.Its parameters are significantly reduced while the detection accuracy is ensured.To solve the problem that the single source ship detection algorithm can hardly resist the interference of cloud layer,sea clutter,island and reef,and cube corner retroreflector,this thesis analyzes the imaging characteristics of visible light and SAR remote sensing images in detail,and then proposes a multi-source remote sensing image decision-level fusion target detection algorithm based on migration learning.,which bypasses the weaknesses.Based on the analysis of different transfer learning modes,a joint detection algorithm based on transfer learning is proposed,and decision-level fusion algorithm is used to fuse the target position,length,width and axial information of multi-model output.The results show that the fusion strategy can be used to improve the anti-interference ability of remote sensing image target detection,the accuracy and recall rate.Finally,through a large number of comparative experiments,the performance of the single model of detection,the training method of transfer learning and the decision-level fusion detection performance are analyzed,and the adaptability of the detection model is tested under different imaging conditions.Porting algorithm is completed on the GPU development board JetsonTX2.Meanwhile,based on the performance analysis of the ported algorithm,the tiny version model is proposed to further improve the real-time performance of the algorithm.