Current Issue Cover


摘 要
目的 遥感图像目标检测旨在定位并识别出遥感图像中的感兴趣目标,它是遥感图像处理的核心问题之一。为解决遥感图像目标检测精度较低的问题。方法 在公开的NWPU_VHR-10数据集上进行实验,对数据集中的一部分低质量图片用增强深度超分辨率(EDSR)网络进行超分辨率重构,为训练卷积神经网络提供高质量数据集。本文对原Faster-RCNN网络进行改进:在特征提取网络中加入注意力机制模块以获取更多所需要关注目标的信息,抑制其他无用信息,以适应遥感图像视野范围大导致的背景复杂和小目标问题;并使用弱化的非极大值抑制来适应遥感图像目标旋转;提出利用目标分布之间的互相关对冗余候选框进一步筛选,降低虚警率,以进一步提高检测器性能。结果 为证明本文方法的有效性,分别进行了两组对比实验,第一组为本文所提各模块间的消融实验,结果表明改进后算法比原始Faster-RCNN的检测结果高了12.2%,证明了本文所提各模块的有效性。第二组为本文方法与其他现有方法在NWPU_VHR-10数据集上的对比分析,本文算法平均检测精度达到79.1%,高于其他对比算法。结论 本文使用EDSR对图像进行超分辨处理,并改进Faster-RCNN,提高了算法对遥感图像目标检测中背景复杂,小目标,物体旋转等情况的适应能力。实验结果表明本文算法的平均检测精度得到了提高。
Attention mechanism improves CNN remote sensing image object detection method

lihongyan,lichungeng(School of Information Science and Technology, Dalian Maritime University)

Objective Remote sensing image object detection aims to locate and identify the object of interest in remote sensing images, which is one of the core issues in remote sensing image processing. Object detection in optical remote sensing images is a fundamental but challenging problem in the field of aerial and satellite image analysis, and is an important part of automated extraction of remote sensing information. Object detection in remote sensing images plays an important role in a wide range of applications: it has broad application value in the fields of national defense security, urban construction planning, and disaster monitoring. And in recent years, it has received great attention. The application range of remote sensing image is expanding day by day, which makes the fast and effective remote sensing object detection method have a very broad application prospect.With the rapid development of the platform and sensor technology, the spatial resolution of remote sensing images continues to increase, and the visual difference from natural images is getting smaller and smaller. More and more computer vision methods can be applied to high spatial resolution remote sensing image object recognition, but there are still problems of low detection accuracy and low efficiency. In order to solve the problem of low accuracy of remote sensing image object detection. Method In this paper, an improved convolutional neural network detection method for attention mechanism is proposed and tested on the NWPU_VHR-10 dataset. The data set is a ten-level geospatial object detection data set. Some of the images have low resolution, which affects the experimental results. Therefore, some low-quality images in the data set were reconstructed with enhanced depth super-resolution (EDSR) network in super-resolution, so as to provide high-quality data set for training convolutional neural network. This paper studied how to use Faster-RCNN model for multi-class object recognition. In this paper, In order to adapt to some characteristics of remote sensing images different from natural images. The original Faster-RCNN network was improved as follows: add attention mechanism in feature extraction network module, get attention convolution neural network, for more information, need to focus on the object by the inhibition of other useless information, so as to adapt to the background of the large range of remote sensing image vision led to complex and problem of small targets; The weak non-maximal suppression is used to adapt to the target rotation of remote sensing image. In order to improve detector performance, the cross-correlation between target distributions is used to further screen redundant candidate frames and reduce false alarm rate. Result In order to prove the validity of the method, two set of comparative experiments were carried out. The first set of comparative experiments is the ablation experiment between the four modules mentioned in this paper: attention mechanism module, non-maximal suppression, cross-correlation filtering mechanism, and image super-resolution processing for low-quality images in this paper. Experimental results show that the improved attentional convolutional neural network has higher detection accuracy than the original Faster-RCNN in ten categories. The average detection accuracy was improved by 12.2%.It is proved that all the modules mentioned in this paper are effective in improving the object detection of aerial remote sensing image. Moreover, the attention module added is a lightweight module that hardly increases the computational cost of the network model. So it doesn''t reduce the efficiency of the network. The second set of comparative experiments is the comparison and analysis of the improved attentional convolutional neural network and other existing traditional methods and deep learning methods on the open data set NWPU_VHR-10. The average detection accuracy of this algorithm is 79.1%, which is higher than other comparison algorithms. Conclusion Convolutional neural network has great application potential in remote sensing image object detection, and is a research hotspot at present and in the future. How to better apply convolutional neural network to object detection of aerial remote sensing images has important theoretical significance. In this paper, the enhanced depth super resolution network is used to superresolve some low resolution images in the dataset. The attentional mechanism was proposed to improve the gross-rcnn, so that the algorithm could pay more attention to the interested target region in the image, that is, the extracted features that are more valuable for the current detection task. It improves the adaptability of the algorithm to the complex background, small object caused by wide field of view, and object rotation caused by aerial photography Angle of view in aerial remote sensing image object detection. Experimental results show that the average detection accuracy of the proposed algorithm is improved.