Real-Time Moving Vehicle Detection in Satellite Video Based on Historical Differential Information and Grouping Features
Cong Xu,Hua Qi,Yahui Zheng,Shiyong Peng
DOI: https://doi.org/10.1109/tgrs.2024.3373621
IF: 8.2
2024-03-30
IEEE Transactions on Geoscience and Remote Sensing
Abstract:Due to the characteristics of small size, few color texture features, and blurred object boundaries of moving vehicles in satellite video, traditional methods based on background subtraction or frame difference and deep learning methods based on single-frame images continue to have a high number of false positives (FPs) and false negatives (FNs). In this article, a satellite video moving vehicle detection network, consisting of the historical frame differential module (HFDM) and the improved YOLOv5 algorithm, is proposed. Initially, a multiscene mosaic (MSM) data augmentation strategy is employed to increase the data complexity, thereby improving the detection ability of the model in complex scenes. Then, HFDM is used to extract the motion information in the historical frames and fuse the motion features with the current frame to obtain the information on the spatio-temporal feature, which can compensate for the absence of single-frame image features. Finally, grouping features of multiple consecutive frames are calculated independently using grouping convolution, the down-sampling rate of the network model is reduced, large-scale feature maps are used for prediction, group attention (GA) and path aggregation network (PANet) are used instead to improve the grouping feature expression of consecutive frames, named GA-PANet. Experiments indicate that the method in this article achieves better detection performance and meets the real-time requirements on both test datasets I and II. The method can deal with the problems of false alarms caused by weak color texture features and blurry boundaries and achieve state-of-the-art performance in real-time moving vehicle detection in satellite video.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics