| Object detection is one of the classic problems in the field of computer vision.Object detection in aerial images has always been a challenging problem.In addition,object detection in aerial images is widely used in civil infrastructure management and military field such as bridge inspection,disaster management,traffic investigation,topographic survey,and aircraft detection.With the development of effective deep learning algorithms and the advancement of hardware systems,higher accuracy has been achieved in detecting various objects from high-resolution satellite images.Focusing on the latest object detection model based on convolutional neural network,some researches are carried out on object detection algorithms in aerial images based on deep learning in this paper.The main contributions are as follows:(1)Investigating the current status of object detection and object detection in aerial images;selecting multiple models of object detection and analyzing their advantages and disadvantages;and realizing automatic detection in aerial images based on the YOLOv5 model by processing multiple aerial image datasets.(2)Considering the sensitivity of the pre-defined anchors of the YOLOv5 model to the dataset,the K-means algorithm is used to cluster the dataset in the experiment to obtain the initial anchors suitable for the dataset,which proves to improve the accuracy of the model.Furthermore,taking into account the image quality of the aerial image dataset,the training dataset of the aerial image dataset is expanded through the mosaic method,cropping,zooming and other data enhancement methods to obtain a more accurate model.(3)The above optimized YOLOv5 model is applied to the NWPU VHR-10 dataset,the RSOD dataset and the DOTA dataset.Experimental results show that the algorithm can detect targets more accurately while ensuring speed.It shows that the improved model in this paper can achieve performance improvement in the task of object detection in aerial images. |
PDF Full Text Request