可见光图像识别开花期苹果叶片SPAD含量估测研究  

Visible light image recognition for estimation of SPAD content in apple leaves at flowering stage

作　　者：候凯耀 李旭[2] 石子琰 邬竞明 HOU Kaiyao;LI Xu;SHI Ziyan;WU Jingming(College of Information Science and Engineering,Xinjiang University of Science and Technology,Korla,Xinjiang 841000;College of Information Engineering,Tarim University,Alar,Xinjiang 843300)

机构地区：[1]新疆科技学院信息科学与工程学院,新疆库尔勒84100 [2]塔里木大学信息工程学院,新疆阿拉尔843300

出　　处：《塔里木大学学报》2025年第1期83-92,共10页Journal of Tarim University

基　　金：塔里木大学校长基金创新团队项目(TDZKCX202306);中国农业大学-塔里木大学联合基金项目(ZNLH202402)。

摘　　要：及时准确地获取果树冠层叶绿素含量信息是农业生产领域密切关注的问题。本研究以新疆南疆重要经济作物苹果树冠层叶片为研究对象,通过获取开花期苹果树冠层叶片的实测SPAD值,结合冠层叶片的可见光图像数据,对颜色特征进行不同通道组合,并探究B、R、G/B、(G-B)/R、(G-B)/(G+B)、(G-B)/(R+G+B)、G-B、R-B、R/B、R/(G+B)、B/(G+R)、(R-B)/G、(R-B)/(R+B)、(G-B)/(G-R)、(R-B)/(R+G+B)、(G-R)/(G+R-B)这16种组合与SPAD值之间的相关性。选择BP神经网络(BPNN)、基于粒子群优化的BP神经网络(PSO-BP)、支持向量回归(SVR)和卷积神经网络(CNN)4种不同算法分别构建预测模型。结果表明,CNN模型的精度最高,训练集R2可以达到0.968,RMSE为1.9969;测试集R2可以达到0.943,RMSE为2.8154。综合分析,可见光图像结合卷积神经网络模型可以监测开花期苹果树冠层叶片叶绿素的含量,为实现无损监测果树叶绿素含量提供参考。Obtaining timely and accurate information on the chlorophyll content in the canopy of fruit trees is a matter of significant concern in the field of production.This study focuses on the canopy leaves of apple trees,a significant economic crop in the Southern Xinjiang region,as the subject of research.By obtaining the actual SPAD values of apple canopy leaves during the flowering period,combined with the visible light image data of the canopy leaves,different channel combinations of color characteristics were explored.The study investigated the correlations between 16 different combinations of color features—B,R,G/B,(G-B)/R,(GB)(/G+B),(G-B)(/R+G+B),G-B,R-B,R/B,R(/G+B),B(/G+R),(R-B)/G,(R-B)(/R+B),(G-B)(/G-R),(R-B)(/R+G+B),and(G-R)(/G+R-B)—and the SPAD values.The study selected four different algorithms to construct predictive models:Backpropagation neural network(BPNN),Particle swarm optimization backpropagation neural network(PSO-BP),Support vector regression(SVR),and Convolutional neural network(CNN).Each of these algorithms was utilized to develop models aimed at forecasting based on the data collected.The results show that the CNN has the highest accuracy,and the training set R2 can reach 0.968 with an RMSE of 1.9969;the test set R2 can reach 0.943 with an RMSE of 2.8154.Upon comprehensive analysis,it is concluded that the combination of visible light imagery with a Convolutional neural network(CNN)model can effectively monitor the chlorophyll content in apple tree canopy leaves during the blooming period.This approach provides a reference for achieving nondestructive monitoring of chlorophyll content in fruit tree leaves.

关 键 词：苹果叶片 SPAD 可见光图像 颜色特征 BP神经网络 卷积神经网络 

分 类 号：S24[农业科学—农业电气化与自动化]