基于密集卷积神经网络的遥感飞机识别
|
摘要
传统的飞机识别方法受模糊、遮挡、噪声以及光照等多种因素的干扰时会降低识别率,且卷积神经网络主要依赖局部特征,却丢失了轮廓特征等重要的全局结构化特征,从而导致算法对于受干扰飞机图像识别效果不佳。因此,基于密集卷积神经网络提出一种结合局部与全局特征的联合监督识别方法,以密集卷积神经网络为基础得到图像特征,通过结合局部特征(卷积神经网络特征)与全局特征(方向梯度直方图特征)进行分类,分类器目标函数使用softmax损失和中心损失联合监督方法。实验结果表明,局部特征与全局特征的结合使算法更加智能化,且损失函数联合监督方法能够实现图像深层特征的类内聚合、类间分散,该算法能有效解决卷积神经网络对受到多种干扰的遥感图像识别率低的问题。
Conventional aircraft identification methods reduce the recognition rate when disturbed by many factors such as blurring, occlusion, noise and illumination, and the convolution neural network relies mainly on the local features while the important global structural features such as contour features are lost, which leads to poor performance of the algorithm in identifying disturbed aircraft images. Therefore, a joint surveillance and recognition method that combines local and global features based on dense convolutional neural network is proposed, and image features are obtained based on dense convolution neural networks and classified by combining local features(convolution neural network features) and global features(direction gradient histogram features). The classifier objective function uses softmax loss and central loss joint monitoring method. The experimental results show that the combination of local features and global features makes the algorithm more intelligent, and the loss function joint supervision method can achieve intra-class aggregation and interclass dispersion of images deep features. This algorithm can effectively solve the problem of low recognition rate of convolution neural networks for remote sensing images that suffer from multiple disturbances.
Conventional aircraft identification methods reduce the recognition rate when disturbed by many factors such as blurring, occlusion, noise and illumination, and the convolution neural network relies mainly on the local features while the important global structural features such as contour features are lost, which leads to poor performance of the algorithm in identifying disturbed aircraft images. Therefore, a joint surveillance and recognition method that combines local and global features based on dense convolutional neural network is proposed, and image features are obtained based on dense convolution neural networks and classified by combining local features(convolution neural network features) and global features(direction gradient histogram features). The classifier objective function uses softmax loss and central loss joint monitoring method. The experimental results show that the combination of local features and global features makes the algorithm more intelligent, and the loss function joint supervision method can achieve intra-class aggregation and interclass dispersion of images deep features. This algorithm can effectively solve the problem of low recognition rate of convolution neural networks for remote sensing images that suffer from multiple disturbances.
引文
[1] Niu Z,Pan J,Lin C,et al.Aircraft recognition in low-resolution SAR imagery using peak feature matching[C]//Proceedings of Internatianal Symposium on Multispectral Image Processing&Pattem Recognition,2015.
[2]朱旭锋,马彩文,刘波.基于特征级融合和支持向量机的飞机识别[J].光电子·激光,2011,22(11):1710-1713.
[3]樊亚军.利用神经网络实现三维飞机目标识别[D].西安:西北工业大学,2005.
[4] Liu G,Sun X,Fu K,et al.Aircraft recognition in highresolution satellite images using coarse-to-fine shape prior[J].IEEE Geoscience&Remote Sensing Letters,2012,10(3):573-577.
[5]赵爱罡,王宏力,杨小冈,等.一种仿射不变的前视红外目标识别方法[J].激光与光电子学进展,2015,52(7):177-184.
[6] Wang D,He X,Wei Z,et al.A method of aircraft image target recognition based on modified PCA features and SVM[C]//Proceedings of the 9th International Conference on Electronic Measurement&Instruments,2009:177-181.
[7]周敏,史振威,丁火平.遥感图像飞机目标分类的卷积神经网络方法[J].中国图象图形学报,2017,22(5):702-708.
[8]周俊宇,赵艳明.卷积神经网络在图像分类和目标检测应用综述[J].计算机工程与应用,2017,53(13):34-41.
[9] Srivastava R K,Greff K,Schmidhuber J.Training very deep networks[C]//Proceedings of NIPS 2015,2015.
[10] He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition,2016:770-778.
[11] Huang Gao,Liu Zhuang,Weinberger K Q.Densely Connected Convolutional Networks[C]//Proceedings of CVPR2017,2017.
[12] Ioffe S,Szegedy C.Batch normalization:Accelerating deep network training by reducing internal covariate shift[C]//Proceedings of International Conference on Machine Learning,2015.
[13]龙胜春,陆嘉炜.基于HOG-GLRLM特征的肠癌病理图片分类[J].计算机工程与应用,2017,53(5):197-201.
[14] Mohan A,Papageorgiou C,Poggio T.Example-based object detection in images by components[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,23(4):349-361.
[15] Wen Y,Zhang K,Li Z,et al.A discriminative feature learning approach for deep face recognition[C]//Proceedings of ECCV 2016,2016:499-515.
[16] Glorot X,Bengio Y.Understanding the difficulty of training deep feedforward neural networks[J].Journal of Machine Learning Research,2010,9:249-256.
[17]邵蔚元,郭跃飞.多任务学习及卷积神经网络在人脸识别中的应用[J].计算机工程与应用,2016,52(13):32-37.
[18] Zhang L,Zhang Y.Airport detection and aircraft recognition based on two-layer saliency model in high spatial resolution remote-sensing images[J].IEEE Journal of Selected Topics in Applied Earth Observations&Remote Sensing,2017,10(4):1511-1524.
[2]朱旭锋,马彩文,刘波.基于特征级融合和支持向量机的飞机识别[J].光电子·激光,2011,22(11):1710-1713.
[3]樊亚军.利用神经网络实现三维飞机目标识别[D].西安:西北工业大学,2005.
[4] Liu G,Sun X,Fu K,et al.Aircraft recognition in highresolution satellite images using coarse-to-fine shape prior[J].IEEE Geoscience&Remote Sensing Letters,2012,10(3):573-577.
[5]赵爱罡,王宏力,杨小冈,等.一种仿射不变的前视红外目标识别方法[J].激光与光电子学进展,2015,52(7):177-184.
[6] Wang D,He X,Wei Z,et al.A method of aircraft image target recognition based on modified PCA features and SVM[C]//Proceedings of the 9th International Conference on Electronic Measurement&Instruments,2009:177-181.
[7]周敏,史振威,丁火平.遥感图像飞机目标分类的卷积神经网络方法[J].中国图象图形学报,2017,22(5):702-708.
[8]周俊宇,赵艳明.卷积神经网络在图像分类和目标检测应用综述[J].计算机工程与应用,2017,53(13):34-41.
[9] Srivastava R K,Greff K,Schmidhuber J.Training very deep networks[C]//Proceedings of NIPS 2015,2015.
[10] He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition,2016:770-778.
[11] Huang Gao,Liu Zhuang,Weinberger K Q.Densely Connected Convolutional Networks[C]//Proceedings of CVPR2017,2017.
[12] Ioffe S,Szegedy C.Batch normalization:Accelerating deep network training by reducing internal covariate shift[C]//Proceedings of International Conference on Machine Learning,2015.
[13]龙胜春,陆嘉炜.基于HOG-GLRLM特征的肠癌病理图片分类[J].计算机工程与应用,2017,53(5):197-201.
[14] Mohan A,Papageorgiou C,Poggio T.Example-based object detection in images by components[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,23(4):349-361.
[15] Wen Y,Zhang K,Li Z,et al.A discriminative feature learning approach for deep face recognition[C]//Proceedings of ECCV 2016,2016:499-515.
[16] Glorot X,Bengio Y.Understanding the difficulty of training deep feedforward neural networks[J].Journal of Machine Learning Research,2010,9:249-256.
[17]邵蔚元,郭跃飞.多任务学习及卷积神经网络在人脸识别中的应用[J].计算机工程与应用,2016,52(13):32-37.
[18] Zhang L,Zhang Y.Airport detection and aircraft recognition based on two-layer saliency model in high spatial resolution remote-sensing images[J].IEEE Journal of Selected Topics in Applied Earth Observations&Remote Sensing,2017,10(4):1511-1524.