﻿ 基于全卷积神经网络的空间植物图像快速识别
 计算机系统应用  2018, Vol. 27 Issue (11): 136-141 PDF

1. 中国科学院 空间应用工程与技术中心 中国科学院太空应用重点实验室，北京100094;
2. 中国科学院大学, 北京 100049

Fast Recognition of Space Plants Image Based on Fully Convolutional Networks
FAN Shuai1,2, WANG Xin1,2, YAN Zhen1
1. CAS Key Laboratory of Space Utilization, Technology and Engineering Center for Space Utilization, Chinese Academy of Sciences, Beijing 100094, China;
2. University of Chinese Academy of Sciences, Beijing 100049, China
Abstract: In order to solve the problem of the long-term survival of the astronauts in the space station, the research of space plants becomes more and more important. At present, there are some problems in image recognition field, such as the method of the shallow images recognition is difficult to extract hierarchical features of space plant images, and deep convolution neural network has fixed size input and long recognition time. To deal with these problems, a method based on fully convolutional networks is proposed in this study, and the networks have the ability to extract features from the shallow to deep, deep fusion spectrum features, and spatial features to achieve an efficient and accurate representation of the space plants image, so as to achieve fast and accurate recognition of the space plants image.
Key words: space plants     image recognition     fully convolutional networks     feature fusion     fast recognition

1 本文方法

1.1 构建全卷积神经网络结构

 图 1 全卷积神经网络结构

1.2 学习全卷积神经网络参数

 图 2 VGGNet

 图 3 VGG部分训练样本

 图 4 FCN训练样本

1.3 基于全卷积神经网络的空间植物图像快速识别

2 实验结果

 图 5 不同生长周期识别结果

 $pixel\ accuracy=\sum\nolimits_{m}{{{n}_{mm}}}/\sum\nolimits_{m}{\sum\nolimits_{l}{{{n}_{ml}}}}$ (1)
 $mean\ accuracy=\left( 1/N \right)\sum\nolimits_{m}{{{n}_{mm}}}/\sum\nolimits_{l}{{{n}_{ml}}}$ (2)

nml表示l类被识别为m类的数量, $\sum\nolimits_{l}{{{n}_{ml}}}$ l类的像素总数, N代表所要识别的类别数目, 取值N=5.

 图 6 干扰条件下的局部识别结果

 图 7 实验对比结果

3 实验结果分析

4 结论与展望

 [1] 赵青青, 张涛, 郑伟波. 基于空间培养的植物图像处理方法. 传感器与微系统, 2016, 35(10): 46-48. [2] 童广辉, 袁永春, 郑伟波, 等. 空间高等植物培养装置. 空间科学学报, 2016, 36(4): 557-561. DOI:10.11728/cjss2016.04.557 [3] 郑伟波, 张涛, 卢晋人, 等. 空间高等植物培养观察装置及其空间飞行实验. 科学技术与工程, 2007, 7(17): 4403-4407, 4414. DOI:10.3969/j.issn.1671-1815.2007.17.031 [4] Li YS, Tao C, Tan YH, et al. Unsupervised multilayer feature learning for satellite image scene classification. IEEE Geoscience and Remote Sensing Letters, 2016, 13(2): 157-161. DOI:10.1109/LGRS.2015.2503142 [5] 管皓, 薛向阳, 安志勇. 深度学习在视频目标跟踪中的应用进展与展望. 自动化学报, 2016, 42(6): 834-847. [6] Chaib S, Gu YF, Yao HX. An informative feature selection method based on sparse PCA for VHR scene classification. IEEE Geoscience and Remote Sensing Letters, 2016, 13(2): 147-151. DOI:10.1109/LGRS.2015.2501383 [7] 黄启宏. 流形学习方法理论研究及图像中应用[博士学位论文]. 成都: 电子科技大学, 2007. [8] 耿蕾蕾. 基于字典学习算法的遥感图像复原及其应用研究[博士学位论文]. 南京: 南京理工大学, 2016. [9] Hinton GE, Salakhutdinov RR. Reducing the dimensionality of data with neural networks. Science, 2006, 313(5786): 504-507. DOI:10.1126/science.1127647 [10] 余凯, 贾磊, 陈雨强, 等. 深度学习的昨天、今天和明天. 计算机研究与发展, 2013, 50(9): 1799-1804. DOI:10.7544/issn1000-1239.2013.20131180 [11] Ciresan DC, Meier U, Masci J, et al. Flexible, high performance convolutional neural networks for image classification. Proceedings of the 22nd International Joint Conference on Artificial Intelligence. Barcelona, Catalonia, Spain. 2011. 1237–1242. [12] Ren SQ, He KM, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. DOI:10.1109/TPAMI.2016.2577031 [13] Pinheiro PO, Collobert R. Recurrent convolutional neural networks for scene labeling. Proceedings of the 31st International Conference on Machine Learning. Beijing, China. 2014. 82–90. [14] Shelhamer E, Long J, Darrell T. Fully convolutional networks for semantic segmentation. Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston, MA, USA. 2015. 3431–3440.