Research On Oriented Object Detection And Fine-grained Image Classification Based On Remote Sensing Images

With the development of modern remote sensing technology and the improvement of the availability of remote sensing image data,highresolution and high-quality remote sensing images have begun to be used in real scenes to solve related practical problems,especially in the maritime field,the use of high-quality and high-resolution remote sensing images to complete the detection and positioning of specific targets has a vital auxiliary role for maritime management.Considering the relative vacancies and requirements of fine-grained classification target detection in the remote sensing field,a new remote sensing image target detection dataset was proposed in this paper.And a deep learning algorithm suitable for oriented fine-grained classification object detection in remote sensing images was introduced.In terms of dataset construction,a new fine-grained classification ship detection dataset based on high-resolution remote sensing images,named FGSD,was constructed.The dataset collects high-resolution remote sensing images that containing ship samples from multiple large ports around the world.Ship samples were fine categorized and annotated with both horizontal and rotating bounding boxes.To further detailed the information of the dataset,we put forward a new representation method of ships’ orientation.For future research,the dock as a new class was annotated in the dataset.Besides,rich information of images was provided in FGSD,including the source port,resolution and corresponding GoogleEarth’s resolution level of each image.As far as we know,FGSD is the most comprehensive ship detection dataset currently.In terms of oriented fine-grained classification object detection algorithm construction,in response to the need for feature fusion of target multi-scale problems in remote sensing images and fine-grained classification tasks,a neural network model named Multiscale-HBP for fine-grained rotating object detection was proposed in this paper,which combines the advantages of Feature Pyramid Network(FPN)and Hierarchical Bilinear Pooling(HBP).The Multiscale-HBP can be used as a feature extraction network for fine-grained classification tasks and finegrained classification target detection tasks in the field of remote sensing.In addition,to solve the problem of complex target background in remote sensing images,we proposed a new supervised hybrid attention module SBAM,experiments have verified that it can alleviate the positioning and classification problems caused by complex backgrounds to a certain extent.Combining the advantages of proposed Multiscale-HBP module and the supervised attention module SBAM,an algorithm model suitable for fine-grained classification object detection in remote sensing images was build,and we further introduce ArcFaceLoss to improve the performance of the model’s classification branch.The final experiment verifies the good performance of the model in the fine-grained classification target detection task of remote sensing images.