网络层数 Network layer | 类型 Type | 滤波器 Filter | 输入尺寸 Input size/pixel |
1 | 卷积层1 Convolution layer 1 | 3 × 3 | 299 × 299 × 3 |
2 | 卷积层2 Convolution layer 2 | 3 × 3 | 149 × 149 × 32 |
3 | 卷积层3 Convolution layer 3 | 3 × 3 | 147 × 147 × 32 |
4 | 卷积层1 Convolution layer 1 | 3 × 3 | 147 × 147 × 64 |
5 | 卷积层4 Convolution layer 4 | 3 × 3 | 73 × 73 × 64 |
6 | 卷积层5 Convolution layer 5 | 3 × 3 | 71 × 71 × 80 |
7 | 卷积层6 Convolution layer 6 | 3 × 3 | 35 × 35 × 192 |
8 | Inception模块组 Module Group | 3 | 35 × 35 × 288 |
9 | Inception模块组 Module Group | 5 | 17 × 17 × 768 |
10 | Inception模块组 Module Group | 3 | 8 × 8 × 1280 |
11 | 池化层2 Pool Layer 2 | 8 × 8 | 8 × 8 × 2048 |
12 | 线性 Linear function | Logits | 1 × 1 × 2048 |
13 | Softmax | Classifier | 1 × 1 × 1000 |

Citation: GAO H Y, GAO X H, FENG Q S, LI W L, LU Z, LIANG T G. Approach to plant species identification in natural grasslands based on deep learning. Pratacultural Science, 2020, 37(9): 1931-1939 doi:

基于深度学习的天然草地植物物种识别方法
-
关键词:
- 深度学习 /
- 天然草地 /
- 植物识别 /
- TensorFlow /
- Inception V3
English
Approach to plant species identification in natural grasslands based on deep learning
-
Key words:
- deep learning /
- natural grassland /
- plant recognition /
- TensorFlow /
- Inception V3
-
-
-
[1]
阚江明, 王怡萱, 杨晓微, 冷萃. 基于叶片图像的植物识别方法[J]. 科技导报科技导报, 2010, 28(23): 81-85.
KAN J M, WANG Y X, YANG X W, LENG C. Plant recognition method based on leaf images[J]. Science & Technology ReviewScience & Technology Review, 2010, 28(23): 81-85. -
[2]
王敬轩, 冯全, 王宇通, 邵新庆. 基于图像识别技术的豆科牧草分类研究[J]. 草地学报草地学报, 2010, 18(1): 37-41. doi:
WANG J X, FENG Q, WANG Y T, SHAO X Q. Study on classification for leguminous forage based on image recognition technology[J]. Acta Agrestia SinicaActa Agrestia Sinica, 2010, 18(1): 37-41. doi: -
[3]
王礼, 洪祖兵, 方陆明, 陈珣, 吴超. 基于iOS系统的观赏植物识别[J]. 浙江农林大学学报浙江农林大学学报, 2018, 35(5): 900-907. doi:
WANG L, HONG Z B, FANG L M, CHEN X, WU C. IOS-based recognition of ornamental plants[J]. Journal of Zhejiang A & F UniversityJournal of Zhejiang A & F University, 2018, 35(5): 900-907. doi: -
[4]
钱亮亮. 基于随机森林的植物叶片识别应用[J]. 现代计算机现代计算机, 2019, (29): 44-47, 51. doi:
QIAN L L. Application of plant leaf recognition based on random forest[J]. Modern ComputerModern Computer, 2019, (29): 44-47, 51. doi: -
[5]
胡直峰. 植物图像识别方法研究及实现. 杭州: 浙江大学硕士学位论文, 2017.
亚博滚球 HU Z F. Research and implementation of plant images recognition methods. Master Thesis. Hangzhou: Zhejiang University, 2017. -
[6]
HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. ScienceScience, 2006, 313(): 504-. doi:
-
[7]
亚博滚球 krizhevsky a, sutskever i, hinton g e. imagenet classification with deep convolutional neural networks. //international conference on neural information processing systems. curran associates inc, 2012: 1097-1105.
-
[8]
段萌. 基于卷积神经网络的图像识别方法研究. 郑州: 郑州大学硕士学位论文, 2017.
亚博滚球 DUAN M. The research on the mothed of image recognition based on convolutional neural networks. Master Thesis. Zhengzhou: Zhengzhou University, 2017. -
[9]
杭州大拿科技公司. 形色. (2017-11-11)[2018-04-20]. .
亚博滚球 DANA. Xingse. (2017-11-11)[2018-04-20]. . -
[10]
北京联星时代科技有限公司. 花伴侣. (2018-04-11)[2018-04-20]. .
Robusoft (Beijing) CO.,LTD. aiPlants. (2018-04-11)[2018-04-20]. . -
[11]
亚博滚球 microsoft corporation. the flower recognition. (2015-12-05)[2018-04-20]. .
-
[12]
袁培森, 黎薇, 任守纲, 徐焕良. 基于卷积神经网络的菊花花型和品种识别[J]. 农业工程学报农业工程学报, 2018, 34(5): 160-166.
YUAN P S, LI W, REN S G, XU H L. Recognition for flower type and variety of chrysanthemum with convolutional neural network[J]. Transactions of the Chinese Society of Agricultural EngineeringTransactions of the Chinese Society of Agricultural Engineering亚博滚球, 2018, 34(5): 160-166. -
[13]
曹中奇, 刘路路, 康孝岩, 张爱武, 柴沙驼. 一种基于卷积神经网络的草地植物识别方法[J]. 人工智能与机器人研究人工智能与机器人研究, 2018, 7(3): 135-146. doi:
CAO Z Q, LIU L L, KANG X Y, ZHANG A W, CHAI S T. Grassland plant identification method based on convolutional neural network[J]. Artificial Intelligence and Robotics ResearchArtificial Intelligence and Robotics Research, 2018, 7(3): 135-146. doi: -
[14]
袁银, 王东斌, 刘永金. 基于深度学习的植物图像识别方法研究[J]. 现代农业科技现代农业科技, 2017, (23): 288-290. doi:
YUAN Y, WANG D B, LIU Y J. Research on plant image recognition method based on depth learning[J]. XianDai NongYe KeJiXianDai NongYe KeJi, 2017, (23): 288-290. doi: -
[15]
赵彦辉, 范欣宁, 张建逵, 谢明. 基于DeepLearning4J on Spark深度学习方法在药用植物图像识别中应用初探[J]. 中国中医药图书情报杂志中国中医药图书情报杂志, 2018, 42(5): 24-28.
ZHAO Y H, FAN X N, ZHANG J K, XIE M. Discussion on application of deep learning method in medical plant image recognition based on DeepLearning4J on Spark[J]. Chinese Journal of Library and Information Science for Traditional Chinese MedicineChinese Journal of Library and Information Science for Traditional Chinese Medicine, 2018, 42(5): 24-28. -
[16]
刘洋, 冯全, 王书志. 基于轻量级CNN的植物病害识别方法及移动端应用[J]. 农业工程学报农业工程学报, 2019, 35(17): 194-204. doi:
LIU Y, FENG Q, WANG S Z. Plant disease identification method based on lightweight CNN and mobile application[J]. Transactions of the Chinese Society of Agricultural EngineeringTransactions of the Chinese Society of Agricultural Engineering, 2019, 35(17): 194-204. doi: -
[17]
google. tensorflow. (2014-11-09)[2018-04-20]. .
-
[18]
黄睿, 陆许明, 邬依林. 基于TensorFlow深度学习手写体数字识别及应用[J]. 电子技术应用电子技术应用, 2018, 44(10): 12-16.
HUANG R, LU X M, WU Y L. Handwriting digital recognition and application based on TensorFlow deep learning[J]. Application of Electronic TechniqueApplication of Electronic Technique亚博滚球, 2018, 44(10): 12-16. -
[19]
田晓明. 基于深度学习的目标检测技术研究及其应用. 长沙: 湖南大学硕士学位论文, 2018.
TIAN X M. The research and application of object detection technology based on deep learning. Master Thesis. Changsha: Hunan University, 2018. -
[20]
林王峰. 基于视频图像的船舶水尺自动测量系统的设计与实现. 厦门: 集美大学硕士学位论文, 2017.
亚博滚球 LIN W F. The design and implementation of a Ship’s draft automatica measurement system base on the video image recognition technology. Master Thesis. Xiamen: JiMei University, 2017. -
[21]
韩伟红, 贾焰, 周斌. 大数据分析关键技术与挑战[J]. 信息技术与网络安全信息技术与网络安全, 2018, 37(4): 11-14.
HAN W H, JIA Y, ZHOU B. The key technologies and challenges of big data analysis[J]. Information Techology and Network SecurityInformation Techology and Network Security, 2018, 37(4): 11-14.
-
[1]
-
亚博滚球
表 1 亚博滚球 Inception V3网络结构
Table 1. Inception V3 network architecture
下载: 导出CSV
表 2 主要训练参数设置
Table 2. Main training parameter settings
参数名称 Parameter 含义 Meaning 设置 Parameter setting checkpoint_exclude_scopes 模型不恢复层
Model exclude scopesInceptionV3/Logits, InceptionV3/Auxlogits trainable_scopes 训练范围 Training scopes – max_number_of_epoches 最大训练回合
Maximum training round30 batch_size 批尺寸 Batch_size 64 learning_rate 初始学习率 Initial learning rate 初始值0.01,每3个epoch减半
The initial value is 0.01, reduce by half for every 3 epochsoptimizer 优化器 Optimizer rmsprop weight_decay 权重衰减 Weight decay 0.00004 下载: 导出CSV
表 3 亚博滚球 不同科植物物种识别准确率
Table 3. 亚博滚球 Accuracy of plant species identification in different families
科名
Family name物种数量
Number of species准确率
Accuracy/%科名
Family name物种数量
Number of species准确率
Accuracy/%菊科 Compositae 55 90.6 伞形科 Umbelliferae 14 80.0 禾本科 Gramineae 39 89.2 毛茛科 Ranunculaceae 14 82.9 豆科 Leguminosae 26 89.2 藜科 Chenopodiaceae 12 88.3 蔷薇科 Rosaceae 17 91.8 蓼科 Polygonaceae 10 84.0 唇形科 Labiatae 16 90.0 龙胆科 Gentianaceae 8 90.0 玄参科 Scrophulariaceae 15 91.1 百合科 Liliaceae 6 90.0 下载: 导出CSV
表 4 亚博滚球 不同地区植物物种识别准确率
Table 4. 亚博滚球 Accuracy of plant species identification in different areas
省
Province物种数量
Number of species准确率
Accuracy/%甘肃 Gansu 195 88.80 四川 Sichuan 72 89.90 内蒙古 Inner Mongolia 115 89.10 下载: 导出CSV
表 5 亚博滚球 不同草地类型植物物种识别准确率
Table 5. Accuracy of plant species identification in different rangeland types
草地类型
Rangeland types物种数量
Number of species准确率
Accuracy/%高寒草甸类
Alpine meadow107 89.7 山地草甸类
Mountain meadow116 89.3 温性草原类
Temperate grassland69 88.6 温性荒漠草原类
Temperate desert grassland21 89.6 下载: 导出CSV
表 6 几种植物识别系统的识别结果
Table 6. 亚博滚球 Identification results of several plant identification systems
识别软件
Identification system物种数量Number of species 识别准确率Accuracy/% 天然草地植物识别模型 Natural grassland plant identification model 293 89.6 形色 Xingse 176 60.1 花伴侣 aiPlants 209 71.3 微软识花 The flower recognition 29 9.9 拍照识花 Paizhaoshihua 45 15.4 本表的识别结果是识别系统给出的第一个结果,即得分最高的结果。
In this table, the identification result is the first result given by the identification systems.下载: 导出CSV
-