基于卷积神经网络的无人机图像自动识别算法

Automatic Recognition Algorithm of UAV Image Based on Convolutional Neural Network

破译无人机原始图像的效率与情报支援的速度是密切相关的,但目前针对合成孔径图像与光学图像的自动识别算法还不够成熟,存在模型较大、识别率较低等问题。从提高模型识别率、模型轻量化入手,提出一种可以有效识别合成孔径雷达(Synthetic Aperture Radar,SAR)图像与光学遥感图像的轻量化卷积神经网络算法。首先对残差收缩网络进行改进,构建特征提取模块,用自适应K值的一维卷积取代全连接层,并在网络中加入空间注意力,提高阈值提取效率;然后用特征提取模块构建模型,并用MSTAR(Maing and Stationany Target Acquistion and Recognition)数据集与UC Merced Land-Use Data Set、SIRI-WHU两类光学遥感图像测试模型性能,实验显示模型是有效的。

The efficiency of deciphering the original image of the UAV is closely related to the speed of intelligence support.However,the current automatic recognition algorithms for synthetic aperture images and optical images are not mature enough,and there are problems such as large models and low recognition rates.Therefore,to improve the model recognition rate and make the model lightweight,a lightweight convolutional neural network algorithm that can effectively identify SAR images and optical remote sensing images is proposed.Firstly,the Residual Shrinkage Network is improved,a feature extraction module is constructed,the fully connected layer is replaced with a one-dimensional convolution with an adaptive K value,and spatial attention is added to the network to improve the threshold extraction efficiency.After that,the feature extraction module is used to build the model,and the MSTAR data set,UC Merced Land-Use Data Set,SIRI-WHU two types of optical remote sensing images are used to detect the effect of the model.Experiment shows that the model is effective.