Research On Remote Sensing Aircraft Classification And Recognition Based On Deep Learning

In recent years,with China’s attention to the field of computer vision and remote sensing satellite technology,optical remote sensing images can be more easily obtained,and the resolution of the avail able optical remote sensing images has been greatly improved,and the data obtained by optical remote sensing images has become more convenient and rich.As the speed of computer computing continues to rise,deep learning,which relies on the speed of computers,has begun to be applied by some researchers in the field of object detection.If deep learning is used for object detection,classification and even tracking,there is no need to build an accurate mathematical model.Object detection and recognition in highresolution optical remote sensing images can detect the position and category of objects in remote sensing images.In the recognition of ground objects in remote sensing images,aircraft is a typical feature in optical remote sensing images.In the detection and classification of aircraft targets in optical remote sensing images,it is of great application value to accurately and quickly detect the position,aircraft type and number of aircraft targets from remote sensing images taken by remote sensing satellites.However,due to the complex background of remote sensing images,the detection and classification of aircraft targets will be affected.In view of this,the work of this article is as follows:（1）Firstly,due to the small number of aircraft datasets of remote sensing images publicly available on the network,the existing public datasets will be expanded by inverting,adding noise,and cropping the acquired datasets,and then adding remote sensing image information containing aircraft targets obtained from Gaofen 1 and Gaofen 6 satellites.Label the dataset using LabelImg labeling software.The aircraft target is divided into two categories based on the characteristic points of the aircraft type.The aircraft engine of the first type is located on both sides of the fuselage near the tail wing,and the overall shape of the wing presents a backward trend.The aircraft engine of the second type is a hanging type under the wing,and the overall shape of the wing presents a backward trend.A total of 3 100 optical remote sensing images in PASCAL VOC format were obtained,with a total of 16456 targets.（2）Propose adding the attention CBAM module to the Neck portion of YOLOv5₆.1 network architecture is used to increase the network’s focus on effective features.And then propose that in YOLOv5₆.1,a BiFPN weighted bidirectional feature pyramid feature fusion structure is added to the network structure to optimize the network structure,allowing the network to better perform feature fusion.（3）Finally,the improved YOLOv5+Attention+BN algorithm was trained and detected on a selfmade aircraft classification dataset.The average accuracy of the YOLOv5+Attention+BN algorithm reached 83%,which was 2.1%higher than before the improvement.Through comparative experiments on classic algorithms and ablation experiments on network modules,the visual results can show that the algorithm proposed in this paper has better detection and classification performance compared to previous algorithms.This study can provide corresponding assistance and support for the classification and detection of aircraft types in China.