基于离散余弦变换和深度网络的地貌图像分类
|
摘要
在未知环境中,无人机(UAV)着陆地貌的自动识别和分类有着极其重要的研究意义,传统的自然场景分类利用的是中层和底层特征信息,但是无人机着陆地貌图像场景复杂、信息丰富,需要较准确的高层语义特征表达。提出了一种基于离散余弦变换(DCT)和深度网络的地貌图像分类方法。首先将离散余弦变换能量集中的优势引入到卷积神经网络(CNN)的高效特征表达中,以降低维度和计算复杂度;然后根据地貌图像特点构建了14层的特征学习网络,并改进了卷积神经网络结构;最后将得到的深层特征输入到支持向量机(SVM)中,快速准确地完成图像分类。实验结果表明,该算法降低了数据冗余,使训练时间大幅度减少,可以自动地学习高层语义特征;所提算法提取的特征具有较好的特征表达,有效地提高了图像分类准确率。
In the unknown environment,the automatic identification and classification of unmanned aerial vehicle(UAV)landing landforms are of great significance.The traditional natural scene classification uses the information of the middle-and the low-level features,but the UAV landing landform image has complex scene and rich information,which needs high-level semantic features to express more accurate information.A landform image classification algorithm based on discrete cosine transform(DCT)and deep network is proposed.First,the advantage of DCT energy concentration is introduced into the efficient feature representation of convolutional neural network(CNN)to reduce the dimensionality and computational complexity.Then a 14-layer feature learning network is constructed based on the characteristics of landform image,and the CNN structure is improved.Finally,the deep features are input into the support vector machine(SVM)to complete the image classification quickly and accurately.Experimental results show that the algorithm reduces data redundancy and training time greatly,and can automatically learn high-level semantic features.The features extracted by the proposed algorithm have better feature expressions and effectively improve the image classification accuracy.
In the unknown environment,the automatic identification and classification of unmanned aerial vehicle(UAV)landing landforms are of great significance.The traditional natural scene classification uses the information of the middle-and the low-level features,but the UAV landing landform image has complex scene and rich information,which needs high-level semantic features to express more accurate information.A landform image classification algorithm based on discrete cosine transform(DCT)and deep network is proposed.First,the advantage of DCT energy concentration is introduced into the efficient feature representation of convolutional neural network(CNN)to reduce the dimensionality and computational complexity.Then a 14-layer feature learning network is constructed based on the characteristics of landform image,and the CNN structure is improved.Finally,the deep features are input into the support vector machine(SVM)to complete the image classification quickly and accurately.Experimental results show that the algorithm reduces data redundancy and training time greatly,and can automatically learn high-level semantic features.The features extracted by the proposed algorithm have better feature expressions and effectively improve the image classification accuracy.
引文
[1]Hinton G E,Osindero S,Teh Y W.A fast learning algorithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
[2]Szegedy C,Liu W,Jia Y Q,et al.Going deeper with convolutions[C].IEEE Conference on Computer Vision and Pattern Recognition,2015:1-9.
[3]Liu D W,Han L,Han X Y.High spatial resolution remote sensing image classification based on deep learning[J].Acta Optica Sinica,2016,36(4):0428001.刘大伟,韩玲,韩晓勇.基于深度学习的高分辨率遥感影像分类研究[J].光学学报,2016,36(4):0428001.
[4]Fang X,Wang G H,Yang H C,et al.High resolution remote sensing image classification combing with mean-shift segmentation and full convolution neural network[J].Laser&Optoelectronics Progress,2018,55(2):022802.方旭,王光辉,杨化超,等.结合均值漂移分割与全卷积神经网络的高分遥感影像分类[J].激光与光电0620001-8子学进展,2018,55(2):022802.
[5]He X F,Zou Z R,Tao C,et al.Combined saliency with multi-convolutional neural network for high resolution remote sensing scene classification[J].Acta Geodaetica et Cartographica Sinica,2016,45(9):1073-1080.何小飞,邹峥嵘,陶超,等.联合显著性和多层卷积神经网络的高分影像场景分类[J].测绘学报,2016,45(9):1073-1080.
[6]Le Q V,Ranzato M,Monga R,et al.Building highlevel features using large scale unsupervised learning[C].IEEE International Conference on Acoustics,Speech and Signal Processing,2013:8595-8598.
[7]Hinton G E,Salakhutdinov R R.Reducing the dimensionality of data with neural networks[J].Science,2006,313(5786):504-507.
[8]Agarwal A,El-Ghazawi T,El-Askary H,et al.Efficient hierarchical-PCA dimension reduction for hyperspectral imagery[C].IEEE International Symposium on Signal Processing and Information Technology,2008:353-356.
[9]Pan Z J,Rust A G,Bolouri H.Image redundancy reduction for neural network classification using discrete cosine transforms[C].IEEE-INNS-ENNS International Joint Conference on Neural Networks,2000,3:3149-3155.
[10]Dabbaghchian S,Ghaemmaghami M P,Aghagolzadeh A.Feature extraction using discrete cosine transform and discrimination power analysis with a face recognition technology[J].Pattern Recognition,2010,43(4):1431-1440.
[11]Sun J P,Liu J Q.Coal and rock recognition method based on low frequency component characteristics of discrete cosine transform and learning vector quantization[J].Industry&Mine Automation,2015,41(11):1-6.孙继平,刘剑桥.基于离散余弦变换低频分量特征及学习向量量化的煤岩识别方法[J].工矿自动化,2015,41(11):1-6.
[12]Zou X Y,Xu X M,Qing C M,et al.High speed deep networks based on Discrete Cosine Transformation[C].IEEE International Conference on Image Processing,2014:5921-5925.
[13]Yan Z Z,Liu J J.Improved image compression coding method based on discrete cosine transform[J].Computer Technology and Development,2016(1):147-149.严珍珍,刘建军.基于离散余弦变换的图像压缩编码方法及改进[J].计算机技术与发展,2016(1):147-149.
[14]Dai C N.SVM visual classification based on weighted feature of genetic algorithm[C].IEEE Sixth International Conference on Intelligent Systems Design and Engineering Applications,2015:786-789.
[15]Chen S Z,Tian Y L.Pyramid of spatial relatons for scene-level land use classification[J].IEEE Transactions on Geoscience and Remote Sensing,2014,53(4):1947-1957.
[16]Xu S H,Mu X D,Zhao P,et al.Scene classification of remote sensing image based on multi-scale feature and deep neural network[J].Acta Geodaetica et Cartographica Sinica,2016,45(7):834-840.许夙晖,慕晓冬,赵鹏,等.利用多尺度特征与深度网络对遥感影像进行场景分类[J].测绘学报,2016,45(7):834-840.
[2]Szegedy C,Liu W,Jia Y Q,et al.Going deeper with convolutions[C].IEEE Conference on Computer Vision and Pattern Recognition,2015:1-9.
[3]Liu D W,Han L,Han X Y.High spatial resolution remote sensing image classification based on deep learning[J].Acta Optica Sinica,2016,36(4):0428001.刘大伟,韩玲,韩晓勇.基于深度学习的高分辨率遥感影像分类研究[J].光学学报,2016,36(4):0428001.
[4]Fang X,Wang G H,Yang H C,et al.High resolution remote sensing image classification combing with mean-shift segmentation and full convolution neural network[J].Laser&Optoelectronics Progress,2018,55(2):022802.方旭,王光辉,杨化超,等.结合均值漂移分割与全卷积神经网络的高分遥感影像分类[J].激光与光电0620001-8子学进展,2018,55(2):022802.
[5]He X F,Zou Z R,Tao C,et al.Combined saliency with multi-convolutional neural network for high resolution remote sensing scene classification[J].Acta Geodaetica et Cartographica Sinica,2016,45(9):1073-1080.何小飞,邹峥嵘,陶超,等.联合显著性和多层卷积神经网络的高分影像场景分类[J].测绘学报,2016,45(9):1073-1080.
[6]Le Q V,Ranzato M,Monga R,et al.Building highlevel features using large scale unsupervised learning[C].IEEE International Conference on Acoustics,Speech and Signal Processing,2013:8595-8598.
[7]Hinton G E,Salakhutdinov R R.Reducing the dimensionality of data with neural networks[J].Science,2006,313(5786):504-507.
[8]Agarwal A,El-Ghazawi T,El-Askary H,et al.Efficient hierarchical-PCA dimension reduction for hyperspectral imagery[C].IEEE International Symposium on Signal Processing and Information Technology,2008:353-356.
[9]Pan Z J,Rust A G,Bolouri H.Image redundancy reduction for neural network classification using discrete cosine transforms[C].IEEE-INNS-ENNS International Joint Conference on Neural Networks,2000,3:3149-3155.
[10]Dabbaghchian S,Ghaemmaghami M P,Aghagolzadeh A.Feature extraction using discrete cosine transform and discrimination power analysis with a face recognition technology[J].Pattern Recognition,2010,43(4):1431-1440.
[11]Sun J P,Liu J Q.Coal and rock recognition method based on low frequency component characteristics of discrete cosine transform and learning vector quantization[J].Industry&Mine Automation,2015,41(11):1-6.孙继平,刘剑桥.基于离散余弦变换低频分量特征及学习向量量化的煤岩识别方法[J].工矿自动化,2015,41(11):1-6.
[12]Zou X Y,Xu X M,Qing C M,et al.High speed deep networks based on Discrete Cosine Transformation[C].IEEE International Conference on Image Processing,2014:5921-5925.
[13]Yan Z Z,Liu J J.Improved image compression coding method based on discrete cosine transform[J].Computer Technology and Development,2016(1):147-149.严珍珍,刘建军.基于离散余弦变换的图像压缩编码方法及改进[J].计算机技术与发展,2016(1):147-149.
[14]Dai C N.SVM visual classification based on weighted feature of genetic algorithm[C].IEEE Sixth International Conference on Intelligent Systems Design and Engineering Applications,2015:786-789.
[15]Chen S Z,Tian Y L.Pyramid of spatial relatons for scene-level land use classification[J].IEEE Transactions on Geoscience and Remote Sensing,2014,53(4):1947-1957.
[16]Xu S H,Mu X D,Zhao P,et al.Scene classification of remote sensing image based on multi-scale feature and deep neural network[J].Acta Geodaetica et Cartographica Sinica,2016,45(7):834-840.许夙晖,慕晓冬,赵鹏,等.利用多尺度特征与深度网络对遥感影像进行场景分类[J].测绘学报,2016,45(7):834-840.