A Research On Object Detection In Remote Sensing Images Based On Multi-scale And Multi-attention

Object detection is an extremely important research direction in the field of image processing.Ground object detection,based on remote sensing satellite imagery,provides the groundwork of numerous applications,so the detection accuracy is of vital importance.It is of great significance in disaster detection and military reconnaissance.But the background of remote sensing images is complex,the object size is various,and there are many small objects.Most of the current methods based on convolutional neural networks have serious information loss,and it is difficult for the network to accurately notice the key areas.In view of the above problems,an object detection method(MA-FPN)based on multi-scale and multi-attention is proposed in this paper,which can effectively make the network pay attention to the location of the object and reduce the loss of small object information.According to the Feature Pyramid Network,the objects of different sizes are predicted by feature maps of different scales,we firstly put forward a global spatial attention module,which extracts spatial location-related information from shallow features and fuses it with deep features to enhance the position expression ability of deep features.Besides,the paper provides a pixel feature attention module:the multi-scale convolution kernel is employed to generate the feature map of the same size as the input,as well as channel attention is used to assign weights to each layer of feature maps to obtain pixel-level attention maps with good details.Experiments on three remote sensing datasets of NWPU,RSOD,and DOTA show that the proposed method has obvious advantages in remote sensing object detection: especially in the NWPU dataset,m AP is improved by 3.75% compared with the benchmark network.Hot map shows that the proposed method can accurately capture the object area,which greatly reduces miss detection and false detection.In addition,m AP increased by 1.3%.even in the object detection of natural scenes,which fully proved that the model in this paper has good generalization.