Maple species identification based on leaf hyperspectral imaging data

Abstract

The present study aimed to identify the leaves of Acer campestre L., A. negundo L. and A. saccharinum L. The experiment was carried out under controlled laboratory conditions in 2021. A Cubert UHD-185 camera was used for hyperspectral imaging (HSI). The content of photosynthetic pigments was determined by the spectrophotometric method. HSI and determination of pigments were carried out during the growing season of maples with intervals between experiments of 7–10 days. The time interval in the vegetation of maples was determined, in which the identification of their species by the random forest (RF) method occurs most accurately. Based on the strength of the correlation with photosynthetic pigments, factor loadings of principal component analysis (PCA), results of the t-test and analysis of variance (ANOVA), a group of vegetation indices (VIs) was selected, the values of which are associated with the species characteristics of maples. It has been established that VIs CRI1, Carter4, Datt3, DPI, EVI, Gitelson2, GMI1, Green_NDVI, MTVI, PRI, REP_Li, TCARI, TVI are significant for the identification of A. campestre, A. negundo and A. saccharinum. The OOB estimate of error rate does not exceed 7% when classified by the RF method using selected VIs. When training on the data of Jul 02, the testing error in the time interval from Jul 09 to Aug 24 did not exceed 20%.