基于深度卷积神经网络的遥感影像目标检测

随着遥感影像数据规模的快速扩张,如何高效准确地识别遥感影像中的典型目标成为当前的研究热点。为解决传统遥感影像目标检测方法准确率低的问题,用基于深度卷积神经网络进行遥感影像目标检测,在遥感影像数据集上用基于Faster-RCNN的神经网络模型对VGG16卷积网络进行训练,对输入的遥感影像通过区域推荐网络标注出待检目标的包围框和置信度,实现对遥感影像的目标检测。以飞机和油罐为例,在TensorFlow深度学习框架下实现了数据预处理、网络训练、目标检测等功能,并在当前测试数据集上取得了较高的检测准确率和置信度。该研究成果可应用于遥感影像解译和处理等相关领域。

Remote Sensing Image Object Detection Based on Deep Convolution Neural Network

With the rapid expansion of remote sensing image data, how to identify typical object efficiently and accurately in remote sensing images has become a hot research area. In order to break through the bottleneck of the accuracy of traditional remote sensing image object detection methods, this paper proposes a deep convolution neural network method to detect remote sensing image objects, and uses the Faster-RCNN neural network model to train VGG16 convolution networks on remote sensing image datasets. The input remote sensing image is marked with the bounding box and confidence level of the object to be inspected through the region proposal network, thereby realizing the target detection of the remote sensing image. Taking aircraft and oil tank as examples, this paper realizes data preprocessing, network training, object detection and other processes under the TensorFlow deep learning framework, and achieves relative high detection accuracy and confidence on the current test dataset. The research results of this paper can apply to remote sensing image interpretation and remote sensing image processing.