Object Detection in High-Resolution Remote Sensing Images Based on a Hard-Example-Mining Network

Lei Zhang,Yuehuan Wang,Yang Huo
DOI: https://doi.org/10.1109/tgrs.2020.3038673
IF: 8.2
2021-10-01
IEEE Transactions on Geoscience and Remote Sensing
Abstract:In recent years, object detection in remote sensing images (RSIs) has attracted much attention for its application value. Compared with traditional methods that are based on manually extracted features, deep learning methods have a great advantage for object detection and have been vastly promoted. However, existing deep learning methods leave much to be desired in the field of RSI object detection due to the large-scale range of the objects and the complex image backgrounds in RSIs. Algorithms need to be specially optimized for this situation. To solve this problem, we propose an effective deep learning-based RSI object detection framework called the multiscale hard-example-mining network (MSHEMN), which is composed of three parts. First, we use the existing ResNet-50 for feature extraction. Second, we propose a multiscale region proposal network (MSRPN), which improves the existing top–down pathway feature pyramid architecture of feature pyramid network (FPN) by adding lateral connection block (LCB) and adaptive feature merge (AFM) to extract features that combine high-resolution and strong semantical information. Finally, a hard-example-mining network (HEMN), which is a cascade multistage detection network integrated with a hard example mining strategy, is proposed to make the detection network focus on hard examples during the training phase by changing the input data distribution of each stage. Extensive experiments on the High-Resolution Remote Sensing Detection (HRRSD) data set have shown the effectiveness of our proposed method, which achieves an average precision (AP) of 62.6 on the testing data set.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics
What problem does this paper attempt to address?
The paper aims to address the problem of object detection in high-resolution remote sensing images (RSIs). Specifically, existing deep learning methods have shortcomings when dealing with large-scale objects and complex backgrounds in remote sensing images, leading to suboptimal detection performance. The paper proposes an effective framework called the Multiscale Hard-Example-Mining Network (MSHEMN) to solve these issues. The main contributions of the paper include: 1. Proposing a network architecture based on Convolutional Neural Networks (CNN) named MSHEMN to improve object detection performance in RSIs, particularly for objects of different scales, and to reduce detection errors caused by complex backgrounds. 2. Designing a Multiscale Region Proposal Network (MSRPN) that can extract and fuse multiscale features, thereby obtaining features that combine high resolution and strong semantic information. 3. Introducing a cascade multi-stage detection framework with a hard-example mining mechanism (Hard-Example-Mining Network, HEMN), which focuses on hard examples during training, reducing detection errors caused by complex backgrounds. With these improvements, the proposed model achieves better performance on the High-Resolution Remote Sensing Detection (HRRSD) dataset and outperforms some existing methods.