Abstract:Deep convolutional neural networks (DCNNs) are driving progress in object detection of high-resolution remote sensing images. Region proposal generation, as one of the key steps in object detection, has also become the focus of research. High-resolution remote sensing images usually contain various sizes of objects and complex background, small objects are easy to miss or be mis-identified in object detection. If the recall rate of region proposal of small objects and multi-scale objects can be improved, it will bring an improvement on the performance of the accuracy in object detection. Spatial attention is the ability to focus on local features in images and can improve the learning efficiency of DCNNs. This study proposes a multi-scale spatial attention region proposal network (MSA-RPN) for high-resolution optical remote sensing imagery. The MSA-RPN is an end-to-end deep learning network with a backbone network of ResNet. It deploys three novel modules to fulfill its task. First, the Scale-specific Feature Gate (SFG) focuses on features of objects by processing multi-scale features extracted from the backbone network. Second, the spatial attention-guided model (SAGM) obtains spatial information of objects from the multi-scale attention maps. Third, the Selective Strong Attention Maps Model (SSAMM) adaptively selects sliding windows according to the loss values from the system’s feedback, and sends the windowed samples to the spatial attention decoder. Finally, the candidate regions and their corresponding confidences can be obtained. We evaluate the proposed network in a public dataset LEVIR and compare with several state-of-the-art methods. The proposed MSA-RPN yields a higher recall rate of region proposal generation, especially for small targets in remote sensing images.

SC-RPN: A Strong Correlation Learning Framework for Region Proposal.

C-RPNs: Promoting object detection in real world via a cascade structure of Region Proposal Networks.

An Enhanced Region Proposal Network for Object Detection Using Deep Learning Method.

Rich Features and Precise Localization with Region Proposal Network for Object Detection.

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

Revisiting Faster R-Cnn: A Deeper Look At Region Proposal Network

A Multi-strategy Region Proposal Network.

Cascade RPN: Delving into High-Quality Region Proposal Network with Adaptive Convolution

Domain-Invariant Region Proposal Network for Cross-Domain Detection

Enhancing representation learning by exploiting effective receptive fields for object detection

Research on Low-Resolution Pedestrian Detection Algorithms Based on R-CNN with Targeted Pooling and Proposal

Context Combined Cascaded Region Proposal Network With Dual Loss Function For Accurate Car Detection

RFP-Net: Receptive field-based proposal generation network for object detection

A Multi-Scale Spatial Attention Region Proposal Network for High-Resolution Optical Remote Sensing Imagery

Relation-aware Siamese Region Proposal Network for Visual Object Tracking

Collaborative Training Between Region Proposal Localization and Classification for Domain Adaptive Object Detection

Label-efficient object detection via region proposal network pre-training

Scale Adaptive Proposal Network for Object Detection in Remote Sensing Images

C-FCN: Corners-based Fully Convolutional Network for Visual Object Detection

IoU-uniform R-CNN: Breaking Through the Limitations of RPN

SA-RPN: A Spacial Aware Region Proposal Network for Acne Detection