基于迁移学习的树种识别
|
摘要
使用树干和树叶图像实现树种自动识别,目前深度学习可以有效的解决该类问题,但它需要大量样本做训练才能达到较高的识别精度。当面对有限图像数量时,提出基于迁移学习的方法,把经过预训练的卷积神经网络模型进行迁移,即共享卷积层和池化层的权重参数,对新的网络模型超参数进行调整,并建立一个包含10种共计2 000张树干图像和8种共计1 725张树叶图像的数据库,把图片分为训练集和测试集,分别利用迁移学习、普通深度学习和SVM分类方法进行训练和测试,并将这3种方法作对比。最后,通过构建树干和树叶图像的混淆矩阵对迁移学习进行具体分析与说明。实验结果表明,通过迁移学习得到的树干和树叶最高识别精度分别达到92. 51%和98. 20%,比普通深度学习提高了51. 38%和51. 69%,比SVM分类方法提高了43. 94%和45. 08%。迁移学习比普通深度学习和传统SVM分类方法更适合用于小样本数据集的分类识别,并且显著优于普通深度学习和SVM分类方法。
Objective Tree trunk and leaf images are used to realize automatic identification of tree species. At present,deep learning can effectively solve this kind of problem,but it needs a large number of samples for training to achieve high recognition accuracy.Methods When the number of images is limited,a method based on transfer learning is proposed to migrate the pre-trained convolution neural network model,that is,the weight parameters of convolution and pooling layer. The parameters of the new network are adjusted and a database containing 10 kinds of 2000 tree trunk images and 8 kinds of 1 725 leaf images are established. The images are divided into training sets and test sets,which are trained and tested by transfer learning,general deep learning and SVM classification methods respectively,and the three methods are compared. Results The experimental results show that the maximum recognition accuracy of tree trunks and leaves obtained through transfer learning is 92. 51% and 98. 20%,respectively,which is higher than general deep learning by 51. 38% and 51. 69%,and higher than SVM classification by 43. 94% and 45. 08%. Conclusion Transfer learning is more suitable for classification and recognition of small sample data sets than general deep learning and traditional SVM classification methods,and is significantly better than general deep learning and SVM classification methods.
Objective Tree trunk and leaf images are used to realize automatic identification of tree species. At present,deep learning can effectively solve this kind of problem,but it needs a large number of samples for training to achieve high recognition accuracy.Methods When the number of images is limited,a method based on transfer learning is proposed to migrate the pre-trained convolution neural network model,that is,the weight parameters of convolution and pooling layer. The parameters of the new network are adjusted and a database containing 10 kinds of 2000 tree trunk images and 8 kinds of 1 725 leaf images are established. The images are divided into training sets and test sets,which are trained and tested by transfer learning,general deep learning and SVM classification methods respectively,and the three methods are compared. Results The experimental results show that the maximum recognition accuracy of tree trunks and leaves obtained through transfer learning is 92. 51% and 98. 20%,respectively,which is higher than general deep learning by 51. 38% and 51. 69%,and higher than SVM classification by 43. 94% and 45. 08%. Conclusion Transfer learning is more suitable for classification and recognition of small sample data sets than general deep learning and traditional SVM classification methods,and is significantly better than general deep learning and SVM classification methods.
引文
[1]贾潇然,刘迎涛.树种识别技术的研究进展[J].林业机械与木工设备,2009,37(9):15-19.JIA X R,LIU Y T. Research advances in identification system of timber-producing tree species management and the development trend[J]. Forestry Machinery&Woodworking Equipment,2009,37(9):15-19.
[2]任洪娥,高洁,马岩.我国木材材种识别技术的新进展[J].木材加工机械,2007,18(4):38-41.REN H E,GAO J,MA Y. The newly evolvement of wood recognition technology in China[J]. Wood Processing Machinery,2007,18(4):38-41.
[3] BEECH E,RIVERS M C,OLDFIELD S,et al. Global Tree Search:the first complete global database of tree species and country distributions[J]. Journal of Sustainable Forestry,2017,36(5):454-489.
[4]梁龙,房桂干,吴珽,等.基于支持向量机的近红外特征变量选择算法用于树种快速识别[J].分析测试学报,2016,35(1):101-106.LIANG L,FANG G G,WU H,et al. Fast identification of wood species by near infrared spectroscopy coupling with support vector machine[J]. Journal of Instrumental Analysis,2016,35(1):101-106.
[5]赵鹏超,戚大伟.基于卷积神经网络和树叶纹理的树种识别研究[J].森林工程,2018,34(1):56-59.ZHAO P C,QI D W. Study on tree species identification based on convolution neural network and leaf texture image[J]. Forest Engineering,2018,34(1):56-59.
[6] MOHANTY S P,HUGHES D P,SALATHE M. Using deep learning for image-based plant disease detection[J]. Frontiers in Plant Science,2016,7:1419.
[7] CARPENTIER M,GIGUERE P,GAUDREAULT J. Tree species identification from bark images using convolutional neural networks[J]. IEEE/RSJ International Conference on Intelligent Robots and Systems,2018:1075-1081.
[8]孙俊,谭文军,毛罕平,等.基于改进卷积神经网络的多种植物叶片病害识别[J].农业工程学报,2017,33(19):209-215.SUN J,TAN W J,MAO H P,et al. Recognition of multiple plant leaf diseases based on improved convolutional neural network[J].Transactions of the Chinese Society of Agricultural Engineering,2017,33(19):209-215.
[9]侯宇青阳,全吉成,王宏伟.深度学习发展综述[J].舰船电子工程,2017,37(4):5-9.HOU Y Q Y,QUAN J C,WANG H W. Review of deep learning development[J]. Ship Electronic Engineering,2017,37(4):5-9.
[10]程玉柱,顾权,王众辉,等.基于深度学习的木材缺陷图像检测方法[J].林业机械与木工设备,2018,46(8):33-36.CHENG Y Z,GU Q,WANG Z H,et al. Wood defect image segmentation based on deep learning[J]. Forestry Machinery&Woodworking Equipment,2018,46(8):33-36.
[11]庄福振,罗平,何清,等.迁移学习研究进展[J].软件学报,2015,26(1):26-39.ZHUANG F Z,LUO P,HE Q,et al. Survey on transfer learning research[J]. Journal of Software,2015,26(1):26-39.
[12] LONG M S,ZHU H,WANG J M,et al. Deep transfer learning with joint adaptation networks[J]. International Conference on Machine Learning,2016,26(1):26-39.
[13]周飞燕,金林鹏,董军.卷积神经网络研究综述[J].计算机学报,2017,40(6):1229-1251.ZHOU F Y,JIN L P,DONG J. Review of convolutional neural network[J]. Chinese Journal of Computers,2017,40(6):1229-1251.
[14] HE K M,ZHANG X Y,REN S Q,et al. Deep residual learning for image recognition[J]. Computer Vision and Pattern Recognition,2016:770-778.
[15] HUANG G,LIU Z,MAATEN L V D,et al. Densely connected convolutional networks[J]. Computer Vision and Pattern Recognition,2017:2261-2269.
[16] SIMONYAN K,ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. Computer Science,2014.
[17]庞俊震,淮永建.基于移动端的月季花快速识别方法研究[J].计算机应用与软件,2017,34(8):36-41.PANG J Z,HUAI Y J. Rapid recognition approach for Chinese rose based on mobile phone[J]. Computer Applications&Software,2017,34(8):36-41.
[18]袁培森,黎薇,任守纲,等.基于卷积神经网络的菊花花型和品种识别[J].农业工程学报,2018,34(5):152-158.YUAN P S,LI W,REN S G,et al. Recognition for flower type and variety of chrysanthemum with convolutional neural network[J]. Transactions of the Chinese Society of Agricultural Engineering,2018,34(5):152-158.
[2]任洪娥,高洁,马岩.我国木材材种识别技术的新进展[J].木材加工机械,2007,18(4):38-41.REN H E,GAO J,MA Y. The newly evolvement of wood recognition technology in China[J]. Wood Processing Machinery,2007,18(4):38-41.
[3] BEECH E,RIVERS M C,OLDFIELD S,et al. Global Tree Search:the first complete global database of tree species and country distributions[J]. Journal of Sustainable Forestry,2017,36(5):454-489.
[4]梁龙,房桂干,吴珽,等.基于支持向量机的近红外特征变量选择算法用于树种快速识别[J].分析测试学报,2016,35(1):101-106.LIANG L,FANG G G,WU H,et al. Fast identification of wood species by near infrared spectroscopy coupling with support vector machine[J]. Journal of Instrumental Analysis,2016,35(1):101-106.
[5]赵鹏超,戚大伟.基于卷积神经网络和树叶纹理的树种识别研究[J].森林工程,2018,34(1):56-59.ZHAO P C,QI D W. Study on tree species identification based on convolution neural network and leaf texture image[J]. Forest Engineering,2018,34(1):56-59.
[6] MOHANTY S P,HUGHES D P,SALATHE M. Using deep learning for image-based plant disease detection[J]. Frontiers in Plant Science,2016,7:1419.
[7] CARPENTIER M,GIGUERE P,GAUDREAULT J. Tree species identification from bark images using convolutional neural networks[J]. IEEE/RSJ International Conference on Intelligent Robots and Systems,2018:1075-1081.
[8]孙俊,谭文军,毛罕平,等.基于改进卷积神经网络的多种植物叶片病害识别[J].农业工程学报,2017,33(19):209-215.SUN J,TAN W J,MAO H P,et al. Recognition of multiple plant leaf diseases based on improved convolutional neural network[J].Transactions of the Chinese Society of Agricultural Engineering,2017,33(19):209-215.
[9]侯宇青阳,全吉成,王宏伟.深度学习发展综述[J].舰船电子工程,2017,37(4):5-9.HOU Y Q Y,QUAN J C,WANG H W. Review of deep learning development[J]. Ship Electronic Engineering,2017,37(4):5-9.
[10]程玉柱,顾权,王众辉,等.基于深度学习的木材缺陷图像检测方法[J].林业机械与木工设备,2018,46(8):33-36.CHENG Y Z,GU Q,WANG Z H,et al. Wood defect image segmentation based on deep learning[J]. Forestry Machinery&Woodworking Equipment,2018,46(8):33-36.
[11]庄福振,罗平,何清,等.迁移学习研究进展[J].软件学报,2015,26(1):26-39.ZHUANG F Z,LUO P,HE Q,et al. Survey on transfer learning research[J]. Journal of Software,2015,26(1):26-39.
[12] LONG M S,ZHU H,WANG J M,et al. Deep transfer learning with joint adaptation networks[J]. International Conference on Machine Learning,2016,26(1):26-39.
[13]周飞燕,金林鹏,董军.卷积神经网络研究综述[J].计算机学报,2017,40(6):1229-1251.ZHOU F Y,JIN L P,DONG J. Review of convolutional neural network[J]. Chinese Journal of Computers,2017,40(6):1229-1251.
[14] HE K M,ZHANG X Y,REN S Q,et al. Deep residual learning for image recognition[J]. Computer Vision and Pattern Recognition,2016:770-778.
[15] HUANG G,LIU Z,MAATEN L V D,et al. Densely connected convolutional networks[J]. Computer Vision and Pattern Recognition,2017:2261-2269.
[16] SIMONYAN K,ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. Computer Science,2014.
[17]庞俊震,淮永建.基于移动端的月季花快速识别方法研究[J].计算机应用与软件,2017,34(8):36-41.PANG J Z,HUAI Y J. Rapid recognition approach for Chinese rose based on mobile phone[J]. Computer Applications&Software,2017,34(8):36-41.
[18]袁培森,黎薇,任守纲,等.基于卷积神经网络的菊花花型和品种识别[J].农业工程学报,2018,34(5):152-158.YUAN P S,LI W,REN S G,et al. Recognition for flower type and variety of chrysanthemum with convolutional neural network[J]. Transactions of the Chinese Society of Agricultural Engineering,2018,34(5):152-158.