Measurement Method of Seedling Stage Maize Height Based on Shuffle-ZoeDepth Monocular Depth Estimation
Abstract
Plant height is an important phenotypic indicator for identifying maize germplasm traits and crop vigor, and maize genetic characteristics are obvious at the seedling stage, so accurate measurement of plant height at the seedling stage is of great significance for maize genetic characteristics identification and field management. Aiming at the problem that traditional plant height acquisition methods rely on manual measurement, which is time-consuming and subjective error, an improved ZoeDepth monocular depth estimation model incorporating mixed attention information was proposed. The improved model added the Shuffle Attention module to the various stages in the Decoder module, so that the Decoder module was more able to pay attention to the effective information in all the feature maps in the process of extracting information from the low-resolution feature maps, which enhanced the model’s ability of key information extraction, and could generate more accurate depth maps. In order to verify the effectiveness of the method, the validation was carried out on the NYU-V2 depth dataset, and the results showed that the ARE, RMSE, LG were 0.083, 0.301mm and 0.036, and the accuracy δ under different thresholds of the improved Shuffle-ZoeDepth model were 93.9%, 99.1% and 99.8%, respectively, all of which were better than those of the improved Shuffle-ZoeDepth model on NYU-V2 depth dataset.In addition, the Shuffle-ZoeDepth monocular depth estimation model combined with the maize plant height measurement model was used to complete the measurement of seedling maize plant height, and maize height measurement experiments were carried out by collecting images of seedling maize at different distances, and when the maize height was in the three height intervals of 15~25cm, 25~35cm, and 35~45cm, the AE were respectively 1.41cm, 2.21cm, and 2.08cm, and the PE were 8.41%, 7.54%, and 4.98%, respectively. The experimental results showed that this method can accomplish the accurate measurement of maize plant height at the seedling stage in complex environments using only a single RGB camera with a complex outdoor environment.
Keywords: maize in seedling stage; plant height; monocular depth estimation; measurement method; hybrid attention mechanism
Download Full Text:
PDFReferences
ZHANG Chunxiao, YUAN Ying, LIU Wenguo, et al. Comparative analysis between salt-alkali tolerance in seedling stage and production in the field of maize hybrid [ J ]. Journal of Maize Sciences, 2010, 18(5) ; 14 - 18. (in Chinese)
DING Hanyun, НЕ Xuchen, FANG Fang, et al. Response mechanism of maize in bred linesto cold stressat seedling stage [J]. Seed, 2022, 41 (10) ; 1 - 10. ( in Chinese)
SUN Jianlu, ZENG Yong, LI Shicheng, et al. Effects of drought stress on maize growth, physiological indexes and quality [ J]. Maize Science, 2023, 31(4) ; 91 -98. (in Chinese)
ZHANG Huichun, WANG Guosu, BIAN Liming, et al. Visible camera-based 3D phenotype measurement system and yime- series visual growth model of plant[ J]. Transactions of the Chinese Society for Agricultural Machinery, 2019, 50( 10) ; 197 — 207. (in Chinese)
XING Haoran. Research on key technologies of remote measurement system of corn plant height [ D ]. Hefei; Anhui Agricultural University, 2020. (in Chinese)
LI Hongwei, DONG Wenhao, LI Zehua, et al. Smartphone application-based measurements of stem-base width and plant height in rice seedling [J] . Computers and Electronics in Agriculture, 2022, 198:107022.
QIU Ruicheng, MIAO Yanlong, JI Yuhan, et al. Measurement of individual maize height based on RGB - D camera [ J]. Transactions of the Chinese Society for Agricultural Machinery, 2017, 48(Supp. ): 211 —219. (in Chinese)
WANG Jizhang, GU Rongrong, SUN Li, et al. Non-destructive monitoring of plug seedling growth process based on Kinect camera [J] . Transactions of the Chinese Society for Agricultural Machinery, 2021 , 52(2) ; 227 -235. (in Chinese)
XU Shengyong, LI Lei, FONG Hui, et al. Research on 3D phenotypic high-throughput measurement system of cucumber seedlings based on RGB - D camera [J ]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54(7) ; 204 - 213,281. (in Chinese)
YUAN W, GU X, DAI Z, el al. NeW CKFs: neural window fully-connected crfs for monocular depth estimation J . ArXiv, 2022, abs/2203. 01502.
WU С Y, WANG J, HALL M, et al. Toward practical monocular indoor depth estimation [ C] // IEEE/C VF Conference on Computer Vision Pattern Recognition, 2021 ; 3804 -3814.
BHAT S, B1KKL K, WOFK D, et al. ZoeDepth: zero-shot transfer by combining relative and metric depth[ J]. ArXiv, 2023, abs/2302.12288.
SONG Lei, LI Kong, JIAO Yitao, et al. Method for measuring seedling height based on KesNeXt monocular depth estimation [J] . Transactions of the CSAE, 2022, 38(3) : 155 - 163. (in Chinese)
LONG VAN, GAO Van, ZHANG Guangben. Depth estimation of apple tree in single image using improved HHNet [J]. Transactions of the CSAE,2022, 38(23) ; 122 - 129. (in Chinese)
LIU Z, HU H, LIN Y, et al. Swin Transformer V2; scaling up capacity and resolution [ C] //IEEE/CVF Conference on Computer Vision Pattern Recognition, 2021 ; 11999 - 12009.
LIU Z, LIN Y, CAO Y, et al. Swin Transformer: hierarchical vision transformer using shifted windows [C] // IEEE/CVF International Conference on Computer Vision, 2021 ; 9992 - 10002.
RANFTL K, BOCHKOVSKIY A, KOLTUN V. Vision transformers for dense prediction [С] // IEEF/CVF International Conference on Computer Vision, 2021 : 12159 -12168.
Refbacks
- There are currently no refbacks.
