新疆农业科学 ›› 2024, Vol. 61 ›› Issue (10): 2547-2556.DOI: 10.6048/j.issn.1001-4330.2024.10.022
• 植物保护·土壤肥料·节水灌溉·农业装备工程与机械化·草业 • 上一篇 下一篇
秦叶康阳(), 李嘉欣, 靳瑰丽(
), 刘文昊, 马建, 李文雄, 陈梦甜
收稿日期:
2024-03-11
出版日期:
2024-10-20
发布日期:
2024-11-07
通信作者:
靳瑰丽(1979-),女,河南兰考人,教授,博士,硕士生导师,研究方向为草地资源与生态,(E-mail)jguili@126.com作者简介:
秦叶康阳(2001-),男,江苏沛县人,本科,研究方向为草业科学,(E-mail)2095782457@qq.com
基金资助:
QIN Yekangyang(), LI Jiaxin, JIN Guili(
), LIU Wenhao, MA Jian, LI Wenxiong, CHEN Mentian
Received:
2024-03-11
Published:
2024-10-20
Online:
2024-11-07
Correspondence author:
JIN Guili (1979-), female, from Lankao,Henan,professor, marster supervisor,Ph.D., research direction:grassland resources and ecology, (E-mail) jguili@126.comSupported by:
摘要:
【目的】 基于UAV和CNN ResNet 18参数调节的伊犁绢蒿荒漠草地植物识别性能分析。【方法】 以集中分布在新疆且受退化威胁较大的伊犁绢蒿(Seriphidium transiliense)荒漠草地为对象,利用低空无人机遥感平台搭载多光谱成像仪采集该草地地物信息,选择卷积神经网络(Convolutional Neural Networks,CNN)ResNet 18模型,设置40轮和80轮的2组训练轮数,8、16、32、64的4组批量规模,0.01、0.005、0.001、0.000 5、0.000 1的5组学习率3类超参数,对比分析不同参数设置下的模型分类性能,探究适合伊犁绢蒿荒漠草地群落主要物种识别的最佳参数组合。【结果】 初始模型超参数设置为训练轮数40轮、批量规模8、学习率0.001时,伊犁绢蒿群落总体分类精度为83.65%,伊犁绢蒿种群分类精度为84.21%,角果藜(Ceratocarpus arenarius)种群精度为81.15%;通过调节模型超参数(超参数设置为练轮数40轮、批量规模32、学习率0.000 5),伊犁绢蒿群落总体分类精度为83.73%,伊犁绢蒿种群精度为89.18%,角果藜种群精度为83.78%,较初始模型分别提高了0.08%、4.97%和2.63%。【结论】 通过调节超参数可获得精度高、耗时短、性能稳定的伊犁绢蒿荒漠草地植物识别模型。
中图分类号:
秦叶康阳, 李嘉欣, 靳瑰丽, 刘文昊, 马建, 李文雄, 陈梦甜. 基于UAV和CNN ResNet 18参数调节的伊犁绢蒿荒漠草地植物识别性能分析[J]. 新疆农业科学, 2024, 61(10): 2547-2556.
QIN Yekangyang, LI Jiaxin, JIN Guili, LIU Wenhao, MA Jian, LI Wenxiong, CHEN Mentian. Identification of main plants in desert grassland Seriphidium transiliense based on UAV remote sensing and CNN ResNet 18[J]. Xinjiang Agricultural Sciences, 2024, 61(10): 2547-2556.
超参数调节 Hyper- parameters adjustment | 序号 Number | 训练轮数 Epoch | 批量规模 Batch sizes | 学习率 Learning rates |
---|---|---|---|---|
训练轮数调节 Epoch adjustment | 1 | 40 | 8 | 0.001 |
2 | 80 | 8 | 0.001 | |
批量规模调节 Batch size adjustment | 3 | 40 | 8 | 0.001 |
4 | 40 | 16 | 0.001 | |
5 | 40 | 32 | 0.001 | |
6 | 40 | 64 | 0.001 | |
学习率调节 Learning rate adjustment | 7 | 40 | 8 | 0.01 |
8 | 40 | 8 | 0.005 | |
9 | 40 | 8 | 0.001 | |
10 | 40 | 8 | 0.000 5 | |
11 | 40 | 8 | 0.000 1 |
表1 ResNet 18模型参数设计
Tab.1 ResNet 18 Model Parameter Design Table
超参数调节 Hyper- parameters adjustment | 序号 Number | 训练轮数 Epoch | 批量规模 Batch sizes | 学习率 Learning rates |
---|---|---|---|---|
训练轮数调节 Epoch adjustment | 1 | 40 | 8 | 0.001 |
2 | 80 | 8 | 0.001 | |
批量规模调节 Batch size adjustment | 3 | 40 | 8 | 0.001 |
4 | 40 | 16 | 0.001 | |
5 | 40 | 32 | 0.001 | |
6 | 40 | 64 | 0.001 | |
学习率调节 Learning rate adjustment | 7 | 40 | 8 | 0.01 |
8 | 40 | 8 | 0.005 | |
9 | 40 | 8 | 0.001 | |
10 | 40 | 8 | 0.000 5 | |
11 | 40 | 8 | 0.000 1 |
序号 Number | 训练 轮数 Epoch | 批量 规模 Batch sizes | 学习率 Learning rates | 识别精度 Identification accuracy(%) | |||
---|---|---|---|---|---|---|---|
伊犁绢蒿 Seriphidium transiliense | 角果藜 Ceratocarpus arenarius | 裸地 Bare land | 总体 Overall accuracy | ||||
1 | 40 | 8 | 0.001 | 84.21 | 81.15 | 86.41 | 83.65 |
2 | 80 | 8 | 0.001 | 80.21 | 76.18 | 91.48 | 82.51 |
表2 不同训练轮数识别精度的变化
Tab.2 Changes of identification accuracy of different training rounds
序号 Number | 训练 轮数 Epoch | 批量 规模 Batch sizes | 学习率 Learning rates | 识别精度 Identification accuracy(%) | |||
---|---|---|---|---|---|---|---|
伊犁绢蒿 Seriphidium transiliense | 角果藜 Ceratocarpus arenarius | 裸地 Bare land | 总体 Overall accuracy | ||||
1 | 40 | 8 | 0.001 | 84.21 | 81.15 | 86.41 | 83.65 |
2 | 80 | 8 | 0.001 | 80.21 | 76.18 | 91.48 | 82.51 |
序号 Number | 训练 轮数 Epoch | 批量 规模 Batch sizes | 学习率 Learning rates | 识别精度 Identification accuracy(%) | |||
---|---|---|---|---|---|---|---|
伊犁绢蒿 Seriphidium transiliense | 角果藜 Ceratocarpus arenarius | 裸地 Bare land | 总体 Overall accuracy | ||||
1 | 40 | 8 | 0.001 | 84.21 | 81.15 | 86.41 | 83.65 |
2 | 40 | 16 | 0.001 | 90.90 | 81.17 | 82.40 | 83.90 |
3 | 40 | 32 | 0.001 | 88.88 | 81.48 | 84.27 | 84.19 |
4 | 40 | 64 | 0.001 | 76.51 | 89.10 | 81.95 | 83.22 |
表3 不同批量规模识别精度的变化
Tab.3 Changes of identification accuracy of different batch sizes
序号 Number | 训练 轮数 Epoch | 批量 规模 Batch sizes | 学习率 Learning rates | 识别精度 Identification accuracy(%) | |||
---|---|---|---|---|---|---|---|
伊犁绢蒿 Seriphidium transiliense | 角果藜 Ceratocarpus arenarius | 裸地 Bare land | 总体 Overall accuracy | ||||
1 | 40 | 8 | 0.001 | 84.21 | 81.15 | 86.41 | 83.65 |
2 | 40 | 16 | 0.001 | 90.90 | 81.17 | 82.40 | 83.90 |
3 | 40 | 32 | 0.001 | 88.88 | 81.48 | 84.27 | 84.19 |
4 | 40 | 64 | 0.001 | 76.51 | 89.10 | 81.95 | 83.22 |
序号 Number | 训练 轮数 Epoch | 批量 规模 Batch size | 学习率 Learning rates | 识别精度 Identification accuracy(%) | |||
---|---|---|---|---|---|---|---|
伊犁绢蒿 Seriphidium transiliense | 角果藜 Ceratocarpus arenarius | 裸地Bare land | 总体 Overall accuracy | ||||
1 | 40 | 8 | 0.01 | 87.94 | 74.31 | 84.65 | 81.29 |
2 | 40 | 8 | 0.005 | 86.42 | 78.85 | 83.85 | 82.40 |
3 | 40 | 8 | 0.001 | 84.21 | 81.15 | 86.41 | 83.65 |
4 | 40 | 8 | 0.000 5 | 89.66 | 79.33 | 86.27 | 84.26 |
5 | 40 | 8 | 0.000 1 | 81.37 | 86.12 | 87.10 | 85.12 |
表4 不同学习率识别精度的变化
Tab.4 Changes of identification accuracy of different learning rates
序号 Number | 训练 轮数 Epoch | 批量 规模 Batch size | 学习率 Learning rates | 识别精度 Identification accuracy(%) | |||
---|---|---|---|---|---|---|---|
伊犁绢蒿 Seriphidium transiliense | 角果藜 Ceratocarpus arenarius | 裸地Bare land | 总体 Overall accuracy | ||||
1 | 40 | 8 | 0.01 | 87.94 | 74.31 | 84.65 | 81.29 |
2 | 40 | 8 | 0.005 | 86.42 | 78.85 | 83.85 | 82.40 |
3 | 40 | 8 | 0.001 | 84.21 | 81.15 | 86.41 | 83.65 |
4 | 40 | 8 | 0.000 5 | 89.66 | 79.33 | 86.27 | 84.26 |
5 | 40 | 8 | 0.000 1 | 81.37 | 86.12 | 87.10 | 85.12 |
评价指标 Evaluating indicator | 学习率Learning rates | ||||
---|---|---|---|---|---|
0.01 | 0.005 | 0.001 | 0.0005 | 0.0001 | |
总体分类精度 Overall accuracy(%) | 81.29 | 82.40 | 83.65 | 84.26 | 85.12 |
收敛速度 Convergence rate | 较快 | 较快 | 较快 | 快 | 较快 |
收敛效果 Convergence effect | 好 | 好 | 好 | 好 | 好 |
拟合效果 Fitting results | 较好 | 较好 | 较好 | 好 | 好 |
运行时间 Running time | 283分53秒 | 355分59秒 | 217分32秒 | 204分39秒 | 226分32秒 |
表5 不同学习率模型性能的综合评价
Tab.5 Comprehensive evaluation of model performance with different learning rates
评价指标 Evaluating indicator | 学习率Learning rates | ||||
---|---|---|---|---|---|
0.01 | 0.005 | 0.001 | 0.0005 | 0.0001 | |
总体分类精度 Overall accuracy(%) | 81.29 | 82.40 | 83.65 | 84.26 | 85.12 |
收敛速度 Convergence rate | 较快 | 较快 | 较快 | 快 | 较快 |
收敛效果 Convergence effect | 好 | 好 | 好 | 好 | 好 |
拟合效果 Fitting results | 较好 | 较好 | 较好 | 好 | 好 |
运行时间 Running time | 283分53秒 | 355分59秒 | 217分32秒 | 204分39秒 | 226分32秒 |
图8 不同学习率下模型识别结果可视化 注:其中序号Ⅰ~Ⅴ表示模型学习率设置为0.01、0.005、0.001、0.000 5、0.000 1
Fig. 8 Visualization of model identification results under different learning rates Notes:The serial numbers Ⅰ~Ⅴ indicate that the model learning rate is set to 0.01, 0.005, 0.001, 0.000 5, 0.000 1
[1] |
靳瑰丽, 魏秀红, 范燕敏, 等. 退化伊犁绢蒿荒漠草地群落特征与地面光谱反射率相关性分析[J]. 草地学报, 2017, 25(2): 282-289.
DOI |
JIN Guili, WEI Xiuhong, FAN Yanmin, et al. Correlation between community characteristics and spectral reflectance of the ground on degradedSeriphidium transiliense desert grassland[J]. Acta Agrestia Sinica, 2017, 25(2): 282-289.
DOI |
|
[2] | 武红旗, 范燕敏, 靳瑰丽, 等. 伊犁绢蒿荒漠草地植物光谱特征[J]. 草业科学, 2019, 36(7): 1765-1773. |
WU Hongqi, FAN Yanmin, JIN Guili, et al. Spectral characteristics of main plant inSeriphidium transiliense desert grassland[J]. Pratacultural Science, 2019, 36(7): 1765-1773. | |
[3] |
韩万强, 靳瑰丽, 岳永寰, 等. 基于高光谱成像技术的伊犁绢蒿荒漠草地主要植物识别参数的筛选[J]. 草地学报, 2020, 28(4): 1153-1163.
DOI |
HAN Wanqiang, JIN Guili, YUE Yonghuan, et al. Screening of identification parameters of main plants onSeriphidium transiliense desert frassland based on hyperspectral imaging technology[J]. Acta Agrestia Sinica, 2020, 28(4): 1153-1163.
DOI |
|
[4] | 杨红艳, 杜健民, 王圆, 等. 基于无人机遥感与卷积神经网络的草原物种分类方法[J]. 农业机械学报, 2019, 50(4): 188-195. |
YANG Hongyan, DU Jianmin, WANG Yuan, et al. Classification method of grassland species based on unmanned aerial vehicle remote sensing and convolutional neural network[J]. Transactions of the Chinese Society for Agricultural Machinery, 2019, 50(4): 188-195. | |
[5] |
马建, 刘文昊, 靳瑰丽, 等. 基于CNN和SVM的无人机多光谱遥感草地植物识别[J]. 草地学报, 2022, 30(11): 3165-3174.
DOI |
MA Jian, LIU Wenhao, JIN Guili, et al. Identification of grassland plants by UAV multispectral remote sensing based on CNN and SVM[J]. Acta Agrestia Sinica, 2022, 30(11): 3165-3174.
DOI |
|
[6] | 卢宏涛, 张秦川. 深度卷积神经网络在计算机视觉中的应用研究综述[J]. 数据采集与处理, 2016, 31(1): 1-17. |
LU Hongtao, ZHANG Qinchuan. Applications of deep convolutional neural network in computer vision[J]. Journal of Data Acquisition and Processing, 2016, 31(1): 1-17. | |
[7] | He K M, Zhang X Y, Ren S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA. IEEE, 2016: 770-778. |
[8] | 张燕斌, 杜健民, 王圆, 等. 基于无人机高光谱遥感和3D-ResNet的荒漠草原地物分类[J]. 中国农机化学报, 2022, 43(4): 66-73. |
ZHANG Yanbin, DU Jianmin, WANG Yuan, et al. Terrain classification in desert steppe based on UAV hyperspectral remote sensing and 3D-ResNet[J]. Journal of Chinese Agricultural Mechanization, 2022, 43(4): 66-73. | |
[9] | 王圆, 毕玉革. 基于无人机高光谱的荒漠草原地物精简学习分类模型[J]. 农业机械学报, 2022, 53(11): 236-243. |
WANG Yuan, BI Yuge. Simplified learning classification model based on UAV hyperspectral remote sensing for desert steppe terrain[J]. Transactions of the Chinese Society for Agricultural Machinery, 2022, 53(11): 236-243. | |
[10] | 汪传建, 赵庆展, 马永建, 等. 基于卷积神经网络的无人机遥感农作物分类[J]. 农业机械学报, 2019, 50(11): 161-168. |
WANG Chuanjian, ZHAO Qingzhan, MA Yongjian, et al. Crop identification of drone remote sensing based on convolutional neural network[J]. Transactions of the Chinese Society for Agricultural Machinery, 2019, 50(11): 161-168. | |
[11] | 徐信罗, 陶欢, 李存军, 等. 基于Faster R-CNN的松材线虫病受害木识别与定位[J]. 农业机械学报, 2020, 51(7): 228-236. |
XU Xinluo, TAO Huan, LI Cunjun, et al. Detection and location of pine wilt disease induced dead pine trees based on faster R-CNN[J]. Transactions of the Chinese Society for Agricultural Machinery, 2020, 51(7): 228-236. | |
[12] | Ghiyamat A, Shafri H Z M, Amouzad Mahdiraji G, et al. Hyperspectral discrimination of tree species with different classifications using single- and multiple-endmember[J]. International Journal of Applied Earth Observation and Geoinformation, 2013, 23: 177-191. |
[13] | 史冰莹, 李佳琦, 张磊, 等. 基于CNN的农作物病虫害图像识别模型[J]. 计算机系统应用, 2020, 29(6): 89-96. |
SHI Bingying, LI Jiaqi, ZHANG Lei, et al. Recognition model of crop pests and diseases images based on CNN[J]. Computer Systems & Applications, 2020, 29(6): 89-96. | |
[14] | Arun P V, Buddhiraju K M, Porwal A, et al. CNN based spectral super-resolution of remote sensing images[J]. Signal Processing, 2020, 169: 107394. |
[15] | 李恒凯, 王利娟, 肖松松. 基于多源数据的南方丘陵山地土地利用随机森林分类[J]. 农业工程学报, 2021, 37(7): 244-251. |
LI Hengkai, WANG Lijuan, XIAO Songsong. Random forest classification of land use in hilly and mountaineous areas of Southern China using multi-source remote sensing data[J]. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(7): 244-251. | |
[16] | 陈之射, 尹芳晔. 机器视觉和人工智能的现代化发展分析[J]. 中国新通信, 2023, 25(2): 36-38. |
CHEN Zhishe, YIN Fangye. Analysis of modern development of machine vision and artificial intelligence[J]. China New Telecommunications, 2023, 25(2): 36-38. | |
[17] | 张文景, 蒋泽中, 秦立峰. 基于弱监督下改进的CBAM-ResNet18模型识别苹果多种叶部病害[J]. 智慧农业(中英文), 2023, 5(1): 111-121. |
ZHANG Wenjing, JIANG Zezhong, QIN Lifeng. Identifying multiple apple leaf diseases based on the improved CBAM-ResNet18 model under weak supervision[J]. Smart Agriculture, 2023, 5(1): 111-121.
DOI |
|
[18] | 姜红花, 杨祥海, 丁睿柔, 等. 基于改进ResNet18的苹果叶部病害多分类算法研究[J]. 农业机械学报, 2023, 54(4): 295-303. |
JIANG Honghua, YANG Xianghai, DING Ruirou, et al. Identification of apple leaf diseases based on improved ResNet18[J]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54(4): 295-303. | |
[19] | 牛智有, 于重洋, 吴志陶, 等. 基于改进ResNet18模型的饲料原料种类识别方法[J]. 农业机械学报, 2023, 54(2): 378-385, 402. |
NIU Zhiyou, YU Chongyang, WU Zhitao, et al. Identification of feed raw material type based on improved ResNet18 model[J]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54(2): 378-385, 402. | |
[20] | 高宏元, 高新华, 冯琦胜, 等. 基于深度学习的天然草地植物物种识别方法[J]. 草业科学, 2020, 37(9): 1931-1939. |
GAO Hongyuan, GAO Xinhua, FENG Qisheng, et al. Approach to plant species identification in natural grasslands based on deep learning[J]. Pratacultural Science, 2020, 37(9): 1931-1939. | |
[21] | 梁俊欢, 董峦, 孙宗玖, 等. 基于改进Faster-RCNN模型的无人机影像白喉乌头物种的检测[J]. 新疆农业大学学报, 2022, 45(4): 323-329. |
LIANG Junhuan, DONG Luan, SUN Zongjiu, et al. Detection ofAconitum leucostomum species in UAV images based on improved faster-RCNN model[J]. Journal of Xinjiang Agricultural University, 2022, 45(4): 323-329. | |
[22] | 马永建. 基于CNN的高分辨率遥感影像典型农作物分类方法研究[D]. 石河子: 石河子大学, 2020. |
MA Yongjian. Study on Typical Crops Classification with High-resolution Remote Sensing Images Based on CNN[D]. Shihezi: Shihezi University, 2020. | |
[23] | 王青青, 董能峰, 杨怡, 等. 基于图像处理和深度学习的中药材鉴别关键技术[J]. 自动化与仪器仪表, 2023,(1): 30-35. |
WANG Qingqing, DONG Nengfeng, YANG Yi, et al. Key TCM identification techniques based on image processing and deep learning[J]. Automation & Instrumentation, 2023,(1): 30-35. |
[1] | 吴风扬, 黄文倩, 田喜, 杨育霖. 基于机器视觉的无序堆放西瓜识别与定位技术[J]. 新疆农业科学, 2024, 61(7): 1805-1813. |
[2] | 赵昊楠, 马海燕, 阿斯娅·曼力克, 田聪, 徐俊, 潘竟, 孙宗玖, 郑逢令. 基于深度学习和GEE的作物遥感分类[J]. 新疆农业科学, 2024, 61(11): 2807-2814. |
[3] | 黄星宇, 孙海荣, 杨寒珺, 张凡凡, |
[4] | 李雨晴, 陈燕红, 李永可, 肖天赐, 李清源. 作物害虫图像智能识别方法[J]. 新疆农业科学, 2023, 60(12): 2973-2981. |
[5] | 侯钰荣, 柯梅, 李超, 兰吉勇, 康帅, 靳俊鹏, 魏鹏. 围栏封育和自由放牧对伊犁绢蒿生长规律和内源激素的影响[J]. 新疆农业科学, 2023, 60(11): 2824-2832. |
[6] | 杨征鹤, 杨会民, 喻晨, 陈毅飞, 周欣, 马艳, 王学农. 设施蔬菜自动对靶喷药技术研究现状与分析[J]. 新疆农业科学, 2021, 58(8): 1547-1557. |
[7] | 张伟荣, 温浩军, 谯超凡, 汪光岩. 基于Mask R-CNN的玉米苗与株芯检测方法[J]. 新疆农业科学, 2021, 58(10): 1918-1928. |
[8] | 魏青, 张宝忠, 魏征. 基于无人机多光谱影像的地物识别[J]. 新疆农业科学, 2020, 57(5): 932-939. |
[9] | 韩万强, 靳瑰丽, 岳永寰, 王惠宁, 宫珂, 吴雪儿, 吾鲁帕·阿得尔卡里. 伊犁绢蒿荒漠草地3种主要植物光谱及植被指数改进[J]. 新疆农业科学, 2020, 57(5): 950-957. |
[10] | 雷声渊, 马本学, 王文霞, 罗秀芝, 李玉洁, 戴建国. 基于微调卷积神经网络迁移学习模式下被害棉叶图像的识别[J]. 新疆农业科学, 2019, 56(7): 1294-1302. |
[11] | 杨晶晶,吐尔逊娜依·热依木,张青青,阿马努拉·依明尼亚孜,雪热提江·麦提努日. 放牧强度对伊犁绢蒿荒漠草地植物群落特征的影响[J]. 新疆农业科学, 2018, 55(8): 1542-1550. |
[12] | 刘慧敏, 董乙强, 安沙舟, 杨娇, 阿依敏·波拉提, 关欣. 伊犁绢蒿荒漠群落特征及植物多样性对短期封育的响应[J]. 新疆农业科学, 2018, 55(7): 1319-1325. |
[13] | 范燕敏, 武红旗, 侯艳娜, 艾科拜尔·安外尔, 古丽包斯旦·艾尔肯. 绢蒿荒漠草地两种主要植物的生态化学计量特征分析[J]. 新疆农业科学, 2018, 55(12): 2305-2311. |
[14] | 张小侠, 董乙强, 安沙舟, 李坤. 伊犁绢蒿荒漠植被特征和多样性退化演替分析[J]. 新疆农业科学, 2017, 54(9): 1721-1729. |
[15] | 成湘;夏小伟;安沙舟. 不同封育年限下伊犁绢蒿荒漠草地动态变化[J]. , 2016, 53(8): 1507-1513. |
阅读次数 | ||||||||||||||||||||||||||||||||||||||||||||||||||
全文 31
|
|
|||||||||||||||||||||||||||||||||||||||||||||||||
摘要 121
|
|
|||||||||||||||||||||||||||||||||||||||||||||||||