基于FC-DenseNet的低空航拍光学图像树种识别
|
摘要
使用低空遥感图像进行图像识别为森林调查和监测提供了新的技术契机。基于无人机低空航拍光学图像,以福建省安溪县崩岗区为研究区,建立FC-Dense Net模型进行树种识别。首先,利用Dense模块提取树种图像特征并增强深层网络信息,透过下采样模块降低图像维度,凸显图像的纹理特征和光谱特征;然后,使用上采样模块还原预测图至原始图像大小,并融合浅层Dense模块信息的丰富特征;最后,采用Softmax分类器实现像素分类,完成树种识别。结果显示,基于低空航拍光学图像,FC-Dense Net模型能够准确区分植被与非植被,定位其空间分布特征,其中,FC-Dense Net-103模型的二分类识别精度为92. 1%,表明FC-Dense Net模型加深网络深度后具有较好的识别效果;将植被与非植被细分为13类,FC-Dense Net-103模型的平均识别正确率达到75. 67%。研究结果表明,基于低空航拍光学图像建立的FC-Dense Net模型具有较高的树种分类精度。由于低空航拍光学图像的成本较低,数据获取费用小,时间周期短,可便于森林资源调查和森林树种检测,为深度学习在树种识别领域的应用提供了新思路。
Image recognition based on low-altitude remote sensing imageries provides a new technological opportunity for forest survey and monitoring. In this study,the authors took the permanent gully in Benggang District,Anxi County,Fujian Province,as an instance and constructed the FC-DenseNet to identify tree species based on the low-altitude aerial optical image of UAV. First,the dense module in the FC-DenseNet model can extract the features of spectral images and enhance the information of the deep network,and the transition down block has an impact on reducing the image dimensions and highlighting the texture and spectral features; then,the transition up block can resize the scale of the predicted image to that of the original image,combined with information fusion of the shallow Dense module; finally,the Softmax classifier is used to achieve pixel-level classification so as to complete the tree species recognition. The results are as follows: ①The FC-DenseNet model based on the low-altitude aerial images not only could identify the difference between vegetation and non-vegetation but also could detect the their spatial distribution. The accuracy of the FC-DenseNet-103 model for vegetation and non-vegetation pixels is 92. 1%,and the 103 layers' network layer is the best network layer. ②Tree species are subdivided into 13 categories,and the accuracy of FC-DenseNet-103 model for dominant species reaches 79%. Some conclusions have been reached: The FC-DenseNet model based on low-altitude aerial optical images has a high tree classification accuracy. With the low cost of low-altitude aerial optical imagery,low data acquisition costs and short time cycles,forest resource surveys and forest species detection can be facilitated. The results obtained by the authors provide a new method in the field of tree recognition using deep learning.
Image recognition based on low-altitude remote sensing imageries provides a new technological opportunity for forest survey and monitoring. In this study,the authors took the permanent gully in Benggang District,Anxi County,Fujian Province,as an instance and constructed the FC-DenseNet to identify tree species based on the low-altitude aerial optical image of UAV. First,the dense module in the FC-DenseNet model can extract the features of spectral images and enhance the information of the deep network,and the transition down block has an impact on reducing the image dimensions and highlighting the texture and spectral features; then,the transition up block can resize the scale of the predicted image to that of the original image,combined with information fusion of the shallow Dense module; finally,the Softmax classifier is used to achieve pixel-level classification so as to complete the tree species recognition. The results are as follows: ①The FC-DenseNet model based on the low-altitude aerial images not only could identify the difference between vegetation and non-vegetation but also could detect the their spatial distribution. The accuracy of the FC-DenseNet-103 model for vegetation and non-vegetation pixels is 92. 1%,and the 103 layers' network layer is the best network layer. ②Tree species are subdivided into 13 categories,and the accuracy of FC-DenseNet-103 model for dominant species reaches 79%. Some conclusions have been reached: The FC-DenseNet model based on low-altitude aerial optical images has a high tree classification accuracy. With the low cost of low-altitude aerial optical imagery,low data acquisition costs and short time cycles,forest resource surveys and forest species detection can be facilitated. The results obtained by the authors provide a new method in the field of tree recognition using deep learning.
引文
[1] Dandois J P,Ellis E C. Remote sensing of vegetation structure using computer vision[J]. Remote Sensing,2010,2(4):1157-1176.
[2]张园，陶萍，梁世祥，等．无人机遥感在森林资源调查中的应用[J]．西南林业大学学报，2011,31(3):49-53.Zhang Y,Tao P,Liang S X,et al. Research on application of UAV RS techniques in forest inventories[J]. Journal of Southwest Forestry College,2011,31(3):49-53.
[3]王伟．无人机影像森林信息提取与模型研建[D]．北京:北京林业大学，2015.Wang W. Forest Information Extraction and Model Building[D].Beijing:Beijing Forestry University,2015.
[4] Zarco-Tejada P J,González-Dugo V,Berni J A J. Fluorescence,temperature and narrow-band indices acquired from a UAV platform for water stress detection using a micro-hyperspectral imager and a thermal camera[J]. Remote Sensing of Environment,2012,117(1):322-337.
[5] Zarco-Tejada P J,Guillén-Climent M L,Hernández-Clemente R,et al. Estimating leaf carotenoid content in vineyards using high resolution hyperspectral imagery acquired from an unmanned aerial vehicle(UAV)[J]. Agricultural and Forest Meteorology,2013,171(8):281-294.
[6] Zarco-Tejada P J,Morales A,Testi L,et al. Spatio-temporal patterns of chlorophyll fluorescence and physiological and structural indices acquired from hyperspectral imagery as compared with carbon fluxes measured with eddy covariance[J]. Remote Sensing of Environment,2013,133(12):102-115.
[7] Zarco-Tejada P J,Catalina A,González M R,et al. Relationships between net photosynthesis and steady-state chlorophyll fluorescence retrieved from airborne hyperspectral imagery[J]. Remote Sensing of Environment,2013,136(136):247-258.
[8] Duan S B,Li Z L,Wu H,et al. Inversion of the PROSAIL model to estimate leaf area index of maize,potato and sunflower fields from unmanned aerial vehicle hyperspectral data[J]. International Journal of Applied Earth Observation and Geoinformation,2014,26(2):12-20.
[9] Calderón R,Navas-Cortés J A,Lucena C,et al. High-resolution airborne hyperspectral and thermal imagery for early detection of verticillium wilt of olive using fluorescence,temperature and narrowband spectral indices[J]. Remote Sensing of Environment,2013,139(139):231-245.
[10]周在明，杨燕明，陈本清．基于可见光波段无人机影像的入侵物种互花米草提取研究[J]．亚热带资源与环境学报，2017,12(2):90-92.Zhou Z M,Yang Y M,Chen B Q. Study on the extraction of exotic species Spartina alterniflora from UAV visible images[J]. Journal of Subtropical Resources and Environment,2017,12(2):90-92.
[11]卫智熠．基于卷积神经网络的可见光图像农作物病虫害的检测[D]．哈尔滨:哈尔滨工业大学，2017Wei Z Y. Plant Disease Detection Based on Visible Images and Convolutional Neural Network[D]. Harbin:Harbin Institute of Technology,2017.
[12] Wang D L,Zhang X M,Liu Y Q. Recognition system of leaf images based on neuronal network[J]. Journal of Forestry Research,2006,17(3):243-246.
[13] Wu Q F,Zhou C L,Wang C N. Feature extraction and automatic recognition of plant leaf using artificial neural network[C]//Proceeding of the 2nd International Conference on Computer Science and Education. 2007:47-50.
[14] Backes A R,Casanova D,Bruno O M. A complex network-based approach for boundary shape analysis[J]. Pattern Recognition,2009,42(1):54-67.
[15] Mallah C,Cope J,Orwell J. Plant leaf classification using probabilistic integration of shape,texture and margin features[C]//Signal Processing,Pattern Recognition and Applications,2013,5(1):1-8.
[16]丁雷龙，李强子，杜鑫，等．基于无人机图像颜色指数的植被识别[J]．国土资源遥感，2016,28(1):78-86. doi:10. 6046/gtzyyg. 2016. 01. 12.Ding L L,Li Q Z,Du X,et al. Vegetation extraction method based on color indices from UAV images[J]. Remote Sensing for Land and Resources,2016,28(1):78-86. doi:10. 6046/gtzyyg. 2016.01. 12.
[17]张华，张改改，吴睿．基于GF-1卫星数据的面向对象的民勤绿洲植被分类研究[J]．干旱区地理，2017,40(4):831-838.Zhang H,Zhang G G,Wu R. Object-based vegetable classification based on GF-1 imagery in Minqin Oasis[J]. Arid Land Geography,2017,40(4):831-838.
[18]井然，邓磊，赵文吉，等．基于可见光植被指数的面向对象湿地水生植被提取方法[J]．应用生态学报，2016,27(5):1427-1436.Jing R,Deng L,Zhao W J,et al. Object-oriented aquatic vegetation extracting approach based on visible vegetation indices[J].Chinese Journal of Applied Ecology,2016,27(5):1427-1436.
[19] Krizhevsky A,Sutskever I,Hinton G E. Image Net classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems. 2012,(1):1097-1105.
[20] Wu X,He R,Sun Z N,et al. A light CNN for deep face representation with noisy labels[J]. IEEE Transactions on Information Forensics and Security,2018,13(11):2884-2896.
[21] Collobert R,Weston J,Karlen M,et al. Natural language processing(almost)from scratch[J]. Journal of Machine Learning Research,2011,12(1):2493-2537
[22] Dollár P,Wojek C,Schiele B,et al. Pedestrian detection:A benchmark[C]//Proceeding of 25th IEEE Conference on Computer Vision and Pattern Recognition. IEEE,2009:304-311
[23]张帅，淮永建．基于分层卷积深度学习系统的植物叶片识别研究[J]．北京林业大学学报，2016,38(9):108-115.Zhang S,Huai Y J. Leaf image recognition based on layered convolutions neural network deep learning[J]. Journal of Beijing Forestry University,2016,38(9):108-115.
[24]陈扬洋．基于卷积神经网络的面向对象遥感影像分类方法研究[D]．北京:中国地质大学，2018.Chen Y Y. OBIA Classification of Remote Sensing Image Based on Convolutional Neural Network[D]. Beijing:China University of Geosciences,2018.
[25] Sladojevic S,Arsenovic M,Anderla A,et al. Deep neural networks based recognition of plant diseases by leaf image classification[J].Computational Intelligence and Neuroscience,2016(6):1-11
[26] Brahimi M,Boukhalfa K,Moussaoui A. Deep learning for tomato diseases:Classification and symptoms visualization[J]. Applied Artificial Intelligence,2017:1-17
[27] Amara J,Bouaziz B,Algergawy A. A deep learning-based approach for banana leaf diseases classification[EB/OL]. http://btw2017. informatik. uni-stuttgart. de/slidesandpapers/E1-10/paper_web. pdf.
[28] Long J,Shelhamer E,Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2014,39(4):640-651.
[29] Huang G,Liu Z,Weinberger K Q,et al. Densely connected convolutional networks[C]//Proceeding of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE,2017,1(2):2261-2269
[30]林敬兰，陈志明，黄炎和，等．安溪县崩岗侵蚀空间分布特征探讨[J]．水土保持研究，2009,16(6):63-68.Lin J L,Chen Z M,Huang Y H,et al. Study on the characteristics of spatial distribution of slope disintegration erosion in Anxi County[J]. Research of Soil and Water Conservation,2009,16(6):63-68.
[31]吴征镒．中国植被[M]．北京:科学出版社，1980:16-19.Wu Z Y. Chinese Vegetation[M]. Beijing:Science Press,1980:16-19.
[32] Ioffe S,Szegedy C. Batch normalization:Accelerating deep network training by reducing internal covariate shift[C]//International Conference on Machine Learning. 2015:448-456.
[33] Jégou S,Drozdzal M,Vazquez D,et al. The one hundred layers tiramisu:Fully convolutional densenets for semantic segmentation[C]//Proceeding of the IEEE Conference on Computer Vision and Pattern Recognition Workshops(CVPRW). IEEE,2017:1175-1183.
[2]张园，陶萍，梁世祥，等．无人机遥感在森林资源调查中的应用[J]．西南林业大学学报，2011,31(3):49-53.Zhang Y,Tao P,Liang S X,et al. Research on application of UAV RS techniques in forest inventories[J]. Journal of Southwest Forestry College,2011,31(3):49-53.
[3]王伟．无人机影像森林信息提取与模型研建[D]．北京:北京林业大学，2015.Wang W. Forest Information Extraction and Model Building[D].Beijing:Beijing Forestry University,2015.
[4] Zarco-Tejada P J,González-Dugo V,Berni J A J. Fluorescence,temperature and narrow-band indices acquired from a UAV platform for water stress detection using a micro-hyperspectral imager and a thermal camera[J]. Remote Sensing of Environment,2012,117(1):322-337.
[5] Zarco-Tejada P J,Guillén-Climent M L,Hernández-Clemente R,et al. Estimating leaf carotenoid content in vineyards using high resolution hyperspectral imagery acquired from an unmanned aerial vehicle(UAV)[J]. Agricultural and Forest Meteorology,2013,171(8):281-294.
[6] Zarco-Tejada P J,Morales A,Testi L,et al. Spatio-temporal patterns of chlorophyll fluorescence and physiological and structural indices acquired from hyperspectral imagery as compared with carbon fluxes measured with eddy covariance[J]. Remote Sensing of Environment,2013,133(12):102-115.
[7] Zarco-Tejada P J,Catalina A,González M R,et al. Relationships between net photosynthesis and steady-state chlorophyll fluorescence retrieved from airborne hyperspectral imagery[J]. Remote Sensing of Environment,2013,136(136):247-258.
[8] Duan S B,Li Z L,Wu H,et al. Inversion of the PROSAIL model to estimate leaf area index of maize,potato and sunflower fields from unmanned aerial vehicle hyperspectral data[J]. International Journal of Applied Earth Observation and Geoinformation,2014,26(2):12-20.
[9] Calderón R,Navas-Cortés J A,Lucena C,et al. High-resolution airborne hyperspectral and thermal imagery for early detection of verticillium wilt of olive using fluorescence,temperature and narrowband spectral indices[J]. Remote Sensing of Environment,2013,139(139):231-245.
[10]周在明，杨燕明，陈本清．基于可见光波段无人机影像的入侵物种互花米草提取研究[J]．亚热带资源与环境学报，2017,12(2):90-92.Zhou Z M,Yang Y M,Chen B Q. Study on the extraction of exotic species Spartina alterniflora from UAV visible images[J]. Journal of Subtropical Resources and Environment,2017,12(2):90-92.
[11]卫智熠．基于卷积神经网络的可见光图像农作物病虫害的检测[D]．哈尔滨:哈尔滨工业大学，2017Wei Z Y. Plant Disease Detection Based on Visible Images and Convolutional Neural Network[D]. Harbin:Harbin Institute of Technology,2017.
[12] Wang D L,Zhang X M,Liu Y Q. Recognition system of leaf images based on neuronal network[J]. Journal of Forestry Research,2006,17(3):243-246.
[13] Wu Q F,Zhou C L,Wang C N. Feature extraction and automatic recognition of plant leaf using artificial neural network[C]//Proceeding of the 2nd International Conference on Computer Science and Education. 2007:47-50.
[14] Backes A R,Casanova D,Bruno O M. A complex network-based approach for boundary shape analysis[J]. Pattern Recognition,2009,42(1):54-67.
[15] Mallah C,Cope J,Orwell J. Plant leaf classification using probabilistic integration of shape,texture and margin features[C]//Signal Processing,Pattern Recognition and Applications,2013,5(1):1-8.
[16]丁雷龙，李强子，杜鑫，等．基于无人机图像颜色指数的植被识别[J]．国土资源遥感，2016,28(1):78-86. doi:10. 6046/gtzyyg. 2016. 01. 12.Ding L L,Li Q Z,Du X,et al. Vegetation extraction method based on color indices from UAV images[J]. Remote Sensing for Land and Resources,2016,28(1):78-86. doi:10. 6046/gtzyyg. 2016.01. 12.
[17]张华，张改改，吴睿．基于GF-1卫星数据的面向对象的民勤绿洲植被分类研究[J]．干旱区地理，2017,40(4):831-838.Zhang H,Zhang G G,Wu R. Object-based vegetable classification based on GF-1 imagery in Minqin Oasis[J]. Arid Land Geography,2017,40(4):831-838.
[18]井然，邓磊，赵文吉，等．基于可见光植被指数的面向对象湿地水生植被提取方法[J]．应用生态学报，2016,27(5):1427-1436.Jing R,Deng L,Zhao W J,et al. Object-oriented aquatic vegetation extracting approach based on visible vegetation indices[J].Chinese Journal of Applied Ecology,2016,27(5):1427-1436.
[19] Krizhevsky A,Sutskever I,Hinton G E. Image Net classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems. 2012,(1):1097-1105.
[20] Wu X,He R,Sun Z N,et al. A light CNN for deep face representation with noisy labels[J]. IEEE Transactions on Information Forensics and Security,2018,13(11):2884-2896.
[21] Collobert R,Weston J,Karlen M,et al. Natural language processing(almost)from scratch[J]. Journal of Machine Learning Research,2011,12(1):2493-2537
[22] Dollár P,Wojek C,Schiele B,et al. Pedestrian detection:A benchmark[C]//Proceeding of 25th IEEE Conference on Computer Vision and Pattern Recognition. IEEE,2009:304-311
[23]张帅，淮永建．基于分层卷积深度学习系统的植物叶片识别研究[J]．北京林业大学学报，2016,38(9):108-115.Zhang S,Huai Y J. Leaf image recognition based on layered convolutions neural network deep learning[J]. Journal of Beijing Forestry University,2016,38(9):108-115.
[24]陈扬洋．基于卷积神经网络的面向对象遥感影像分类方法研究[D]．北京:中国地质大学，2018.Chen Y Y. OBIA Classification of Remote Sensing Image Based on Convolutional Neural Network[D]. Beijing:China University of Geosciences,2018.
[25] Sladojevic S,Arsenovic M,Anderla A,et al. Deep neural networks based recognition of plant diseases by leaf image classification[J].Computational Intelligence and Neuroscience,2016(6):1-11
[26] Brahimi M,Boukhalfa K,Moussaoui A. Deep learning for tomato diseases:Classification and symptoms visualization[J]. Applied Artificial Intelligence,2017:1-17
[27] Amara J,Bouaziz B,Algergawy A. A deep learning-based approach for banana leaf diseases classification[EB/OL]. http://btw2017. informatik. uni-stuttgart. de/slidesandpapers/E1-10/paper_web. pdf.
[28] Long J,Shelhamer E,Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2014,39(4):640-651.
[29] Huang G,Liu Z,Weinberger K Q,et al. Densely connected convolutional networks[C]//Proceeding of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE,2017,1(2):2261-2269
[30]林敬兰，陈志明，黄炎和，等．安溪县崩岗侵蚀空间分布特征探讨[J]．水土保持研究，2009,16(6):63-68.Lin J L,Chen Z M,Huang Y H,et al. Study on the characteristics of spatial distribution of slope disintegration erosion in Anxi County[J]. Research of Soil and Water Conservation,2009,16(6):63-68.
[31]吴征镒．中国植被[M]．北京:科学出版社，1980:16-19.Wu Z Y. Chinese Vegetation[M]. Beijing:Science Press,1980:16-19.
[32] Ioffe S,Szegedy C. Batch normalization:Accelerating deep network training by reducing internal covariate shift[C]//International Conference on Machine Learning. 2015:448-456.
[33] Jégou S,Drozdzal M,Vazquez D,et al. The one hundred layers tiramisu:Fully convolutional densenets for semantic segmentation[C]//Proceeding of the IEEE Conference on Computer Vision and Pattern Recognition Workshops(CVPRW). IEEE,2017:1175-1183.