[1]阳 昊,黄 超,刘欣然,等.基于卷积神经网络特征提取的莴苣生长无损监测[J].计算机技术与发展,2023,33(08):137-143.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 020]
 YANG Hao,HUANG Chao,LIU Xin-ran,et al.Convolutional Neural Networksed Feature Extraction for Non-destructive Lettuce Growth Monitoring[J].,2023,33(08):137-143.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 020]
点击复制

基于卷积神经网络特征提取的莴苣生长无损监测()
分享到:

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年08期
页码:
137-143
栏目:
人工智能
出版日期:
2023-08-10

文章信息/Info

Title:
Convolutional Neural Networksed Feature Extraction for Non-destructive Lettuce Growth Monitoring
文章编号:
1673-629X(2023)08-0137-07
作者:
阳 昊1 黄 超12 刘欣然1 王中举1 王 龙12
1. 北京科技大学 计算机与通信工程学院,北京 100083;
2. 北京科技大学 顺德创新学院,广东 佛山 528399
Author(s):
YANG Hao1 HUANG Chao12 LIU Xin-ran1 WANG Zhong-ju1 WANG Long12
1. School of Computer and Communication Engineering,University of Science and Technology Beijing,Beijing 100083,China;
2. Shunde Innovation School,University of Science and Technology Beijing,Foshan 528399,China
关键词:
莴苣无损监测卷积神经网络机器学习特征提取
Keywords:
lettucenon-destructive monitoringconvolutional neural networkmachine learningfeature extraction
分类号:
TP301
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 08. 020
摘要:
基于计算机视觉的莴苣生长无损监测对莴苣的种植管理有重要的意义。 彩色图像包含的纹理、色泽等信息与莴苣生长参数密切相关。 以往的研究包括基于人为设计的特征结合机器学习算法估计生长参数,以及通过卷积网络估计生长参数。 该文设计了一种结合卷积神经网络和机器学习模型的二阶段算法,用于莴苣的生长参数无损估计。 生长参数包括叶片鲜重( LFW)、叶片干重( LDW) 、植株高度(H)、植株直径( D) 和叶面积( LA)。 算法的第一阶段训练卷积网络自动从图像中提取特征,第二阶段基于卷积网络提取的特征,利用集成机器学习算法 Stacking( 随机森林,深度森林) 估计生长参数。 实验结果表明,相比直接使用卷积网络估计,设计的二阶段算法能显著降低误差,在五个生长参数上的归一化均方误差(NMSE) 分别为 2. 25% ,2. 61% ,1. 63% ,0. 84% ,3. 18% , 估计值与真实值的决定系数( R2 ) 为 0. 955 2,0. 957 8,0. 892 1,0. 884 4,0. 936 2。 通过引入深度图,使用 3D 卷积网络从彩色图和深度图的组合中提取特征,高度( H) 的估计准确度能进一步提高( NMSE:1. 27% ,R2 :0. 916 1) 。 表明通过卷积神经网络自动从图片中提取特征并结合集成机器学习算法用于莴苣的生长参数估计是可行的。
Abstract:
Computer vision based non-destructive growth monitoring is essential for lettuce planting management. The texture and coloretc. in lettuce image are?
closely related with its growth traits. Previous studies on image-based traits estimation include machine learning-based methods with hand-crafted features and convolutional neural network ( CNN) -based methods. We propose a two-stage methodconsisting of CNN and machine learning for lettuce growth traits estimation, including leaf fresh weight ( LFW ) , leaf dry weight( LDW) ,height ( H) ,diameter ( D) ,and leaf area ( LA) . In the first stage,CNN is trained to automatically extract feature from images.In the second stage,Stacking regression ( random forest,deep forest) is trained to estimate traits based on extracted features by CNN. Experiment results illustrate that the proposed two stage method outperforms CNN - based method with normalized mean square error( NMSE) of 2. 25% ,2. 61% ,1. 63% ,0. 84% ,3. 18% for five traits respectively,and with R2 values of 0. 955 2,0. 957 8,0. 892 1,0. 884 4,0. 936 2,respectively. For height estimation,the introduction of depth image could further enhance performance with NMSE of1. 27% and R2 of 0. 916 1. The result indicates that CNN based feature extraction is valuable for lettuce growth traits estimation.
更新日期/Last Update: 2023-08-10