文献类型： 外文期刊

作者： Che, Shuai ¹ ; Wu, Lan ¹ ; Wang, Zhen-Dong ¹ ; Tian, Lin ¹ ; Du, Guo-Ying ¹ ; Mao, Yun-Xiang ¹ ;

作者机构： 1.Ocean Univ China, Coll Marine Life Sci, MOE Key Lab Marine Genet & Breeding, Qingdao, Peoples R China

2.Hainan Trop Ocean Univ, MOE Key Lab Utilizat & Conservat Trop Marine Biore, Sanya, Peoples R China

3.Hainan Trop Ocean Univ, Yazhou Bay Innovat Inst, Sanya, Peoples R China

4.Ocean Univ China, Sanya Oceanog Inst, Key Lab Trop Aquat Germplasm Hainan Prov, Sanya, Peoples R China

5.Chinese Acad Fishery Sci, Yellow Sea Fisheries Res Inst, Natl Key Lab Mariculture Biobreeding & Sustainable, Qingdao, Peoples R China

6.Laoshan Lab, Lab Marine Biol & Biotechnol, Qingdao, Peoples R China

关键词： Rhodophyta; Nitrogen content; Macroelements; Phenotyping; Macroalgae; Hyperspectral imaging

期刊名称：JOURNAL OF APPLIED PHYCOLOGY （ 影响因子：3.3； 五年影响因子：3.6 ）

ISSN： 0921-8971

年卷期： 2024 年

页码：

收录情况： SCI

摘要： In the economically important red macroalga Pyropia, carbohydrates and proteins contents are usually used to evaluate the nutritional status and commercial value. As main component elements of carbohydrates and proteins, C, N and S contents can be the suitable indicators. However, efficient and nondestructive estimation of these elements have not been well established in macroalgae, which is definitely needed in high-throughput phenotyping and selective breeding. In the current study, hyperspectral imaging was used to estimate the C, N and S contents in thalli of Pyropia yesoensis. Based on spectral information acquired by two hyperspectral cameras with range of 400 nm to 1700 nm, two machine learning methods, partial least squares regression (PLSR) and support vector machine regression (SVR), were performed to establish prediction models following different preprocessing methods. The result showed that SVR model following Savitzky-Golay (S-G) smoothing preprocessing performed the best for N content with excellent accuracy (R-Test(2) = 0.94, RMSE = 0.331, MAPE = 5.47%, RPD = 3.44). Both of PLSR and SVR models could not yet meet satisfactory prediction for C and S content with R-Test(2) < 0.4 and RPD < 1.5. Further validation on field samples corroborated accuracy and robustness of the optimal estimation model for N. This study demonstrates that hyperspectral imaging is effective for estimating N contents of P.yesoensis thalli. It unveils the capability of hyperspectral imaging with machine learning models for estimating macroelement contents in macroalgae thalli, and offers a convenient, non-destructive, and effective method for the phenotyping and selective breeding of macroalgae.