针对传统遥感图像检测算法中人为干预多、速度慢、检测精度低等问题,提出一种基于深度学习的遥感图像检测方法。采用密集连接的网络结构,充分利用每层网络提取的特征,减少网络推理时间;采用具有更大感受野的扩张块结构;使用扩张块结构和反卷积网络结构将浅层特征图和深层特征图进行信息融合,从而增强遥感图像中多尺度目标的检测能力。实验结果表明,该检测方法具有更高的准确率和更短的检测时间,尤其在小目标物体的检测上表现出更好的性能。

This study proposes a remote sensing image detection method based on deep learning to solve the issues of human intervention, slow speed, and low accuracy associated with the traditional remote sensing image detection algorithm. A dense connected network is considered to completely use the features extracted from each layer and reduce the network inference time. Further, an expanding block structure with a large perceptive field is adopted, and the low- and high-level feature informations of the network are combined based on the expanding block structure and deconvolution network. Thus, the performance of multiscale object detection for remote sensing images is improved. The experimental results denote that the proposed method exhibits high accuracy and short detection time, especially during the detection of small objects.