Phenotypic Detection of Maize Based on Fusion of RGB Image and Multispectral Image Data
Yang Linlin
Liu Rui
Bie Shufan
Li Wenfeng
Shi Jie
Abstract：In order to address the issues of waste time and labor during leaf phenotype collection and inad-equate generalization performance of single-source image detection,this study proposed a deep learning-based algorithm for detecting maize leaf phenotypes(fresh weight,chlorophyll content,leaf area and leaf width)by fusing RGB images and multispectral images.The RGB images and multispectral images of maize leaves were collected at 5 to 7 leaf stage,and semantic segmentation and GrabCut algorithm were used to segment them,re-spectively.A multi-channel input method was employed to input the two types of images into the MobileNetV2 network for training and testing.The results indicated that the effect of maize leaf phenotype detection based on the fusion of RGB and multispectral images outperformed that using single-source data.The average absolute er-rors were 0.161 9,0.110 1,0.166 3 and 0.144 2 respectively for the four phenotypic traits of maize leaves,and the time consumption for each sample was less than 10 ms,meeting expectations.It showed that the proposed maize leaf phenotype detection algorithm based on multisource data fusion exhibited excellent performance and had significant application prospects in early growth monitoring and yield estimation of maize.
Keywords：RGB imagesMultispectral imagesMulti-source data fusionMaize leaf phenotypeImage processingMulti-channel input
Publication Date：2025-10-30
Online Publishing Date：2026-05-22(First online date of this platform, not the publication date of the document)
Pages：8( 158-164,172 )