基于最优区域生成的深度多尺度融合遥感飞机检测方法

遥感图像中典型目标的检测是当前图像处理领域的研究热点,飞机在战场监视、航空管制和交通运输等领域发挥着重要作用。为了提高遥感图像中飞机检测的正确率,提出了一种基于多特征融合的遥感飞机检测方法,将深层特征经过上采样操作后与浅层特征进行融合,解决了遥感飞机目标较小造成的检测困难的问题。首先,对于锚框尺寸和个数由人为确定而造成目标位置检测不准的问题,采用K-均值聚类(K-means)算法对数据集的目标框大小进行聚类分析并获得适合飞机遥感图像的锚框(anchor boxes)个数以及宽高维度;其次,采用上采样的方法扩大感受野,以提高网络对小目标的检测准确率。采用多尺度融合的卷积神经网络,以适应不同尺度目标的检测,最终提出一种基于最优区域生成的深度多尺度融合遥感飞机检测方法。仿真结果表明:与典型的飞机检测方法相比,所提方法在测试集上取得了更高的的检测精度。

Deep Multi-scale Fusion Remote Sensing Aircraft Detection Method Based on Optimal Region Propose

The detection of typical targets in remote sensing images is a hot topic in the field of image processing. Aircraft plays an important role in battlefield surveillance, air traffic control and transportation. In order to improve the accuracy of aircraft detection in remote sensing images, this paper proposes a remote sensing aircraft detection method based on multi-feature fusion. The deep features are merged with the shallow features after up-sampling operation, which solves the problem caused by the small target of remote sensing aircraft. Firstly, because the size and number of anchors are artificially determined, which results in the target detection is not accurate. K-means clustering (k-means) algorithm is used to cluster the size of target frames in the data set and obtain the number of anchors and the dimension of width and height suitable for airplane remote sensing images, In order to improve the detection accuracy of the network for small targets, the method of up-sampling is used to expand the receptive field. In order to adapt to the detection of different scale targets, a multi-scale fusion convolutional neural network is adopted, and a deep multi-scale fusion remote sensing aircraft detection method based on optimal region generation is proposed. The simulation results show that compared with the typical aircraft detection method, the proposed method achieves higher detection accuracy on the test set.