Study On Identification Method Of Healthy Rapeseed (Brassica Napus L.) Seedling Using Hyperspectral Imaging Technology

Nurturing seedlings is a key part of transplanting.In order to ensure the quality of seedlings and provide healthy seedlings suitable for planting,it is necessary to identify the healthy rapeseed seedlings quickly and accurately to meet the needs of large-scale and standardization of modern rapeseed industry.In this paper,rapeseed seedling was taken as the research object with hyperspectral imaging technology used to carry out temperature stress experiment and light stress experiment to study the identification method of healthy rape seedlings.The main research contents are as follows:(1)Proposed a wavelet feature extraction algorithm based on continuous wavelet transform and stepwise discriminant analysis.Aiming at identifying healthy seedlings from those under mild stress,multi-scale continuous wavelet transform was performed on the canopy spectrum,wavelet coefficients sensitive to stress were extracted by one-way ANOVA and SNK test,and a stepwise discriminant analysis method was used to select features to establish CWT-SDA-Fisher discriminant model with the average accuracy of 88.59%±6.90% of temperature stress detection and 77.25%±6.94% of light stress detection.(2)Established multi-feature fusion models for temperature and light stress detection.Aiming at improving stress detection accuracy based only on characteristic wavelengths or wavelet coefficients.A total of 7 features including 554～714 nm AUC,tan(?),1213 nm,1567 nm,w(9,967),w(13,1213),w(7,1567)were selected among characteristic wavelengths,wavelet coefficients and band features extracted by SPA,CWT-SDA,PCF to establish a multi-feature fusion Fisher discriminant model with the average accuracy of 88.68%±7.86% of temperature stress detection,and the best detection accuracy reached 95.56% in the three-leaf stage.Similarly,a total of 4 features including 939～978 nm AUC,tanq,984 nm and 1408 nm were selected to establish a multi-feature fusion light stress detection model with the average accuracy of 88.68%±7.86%.(3)Developed rapeseed seedling’s spectrum and morphological characteristics analysis software.The software includes preprocessing,feature extraction,and feature evolution rule analysis modules,which meets the needs of rapeseed seedling’s feature analysis and stress detection before transplanting.