基于CNN及Bi-LSTM的无人机时序图像定位研究

发布时间：2018-03-23 22:08

本文选题：无人机　切入点：航拍图像　出处：《电光与控制》2017年12期

【摘要】：设计了一个浅层卷积神经网络来代替预训练模型中的全连接层,将预训练网络提取的CNN特征作为图像输入设计好的浅层CNN网络,对比微调预训练模型的方法,能够更好地适应航拍图像定位任务。为进一步提高航拍图像的定位准确率,利用无人机航拍图像时间连续的特点,通过在CNN的分类阶段加入Bi-LSTM网络,使网络在分类时能够以多张图像特征作为判断依据。实验表明,时序图像定位方法定位准确率稳定在0.89左右,对比单张图像定位方法准确率提升5%左右。
[Abstract]:In this paper, a shallow convolution neural network is designed to replace the full connection layer in the pre-training model. The CNN features extracted by the pre-training network are regarded as the shallow CNN network designed for image input, and the methods of fine-tuning the pre-training model are compared. In order to further improve the accuracy of aerial image location, by adding Bi-LSTM network to the classification stage of CNN, the UAV aerial image is continuous in time. The experimental results show that the accuracy of sequential image localization method is about 0.89, and the accuracy of this method is improved by about 5% compared with single image location method.
【作者单位】：空军航空大学;
【基金】：吉林省自然科学基金(20130101069JC)
【分类号】：P23;TP391.41

【相似文献】