Experimental Study Of Rapeseed Recognition Model Base On Measured Spectra

With great advances in information technology, Hyper-spectral Remote Sensing as a rapid mean of monitoring has been used widely in agriculture. It is also a fast and non-destructive method for crop classification. In this study, leaf and canopy spectral reflectance data of rape in three periods (seedling phase, bolting phase and flowering phase) were collected by ASD HandHeld2. This paper analyzed the difference and similarity of varied rape spectral data from varied spectrum characters reflected by varied spectral data of rape. Then, this paper determined the best period for the research of rape spectral based on spectral characteristics of rape in different periods. Finally, the identification model was established by using Mahalanobis distance method, cohesion coefficient analysis, characteristic wavelength index analysis and Principal Component Analysis dimensionality reduction.Specific results show as below:(1)There is the similar spectral character existing between different spectral curves of rape canopy. These spectral curves look roughly similar to the original spectral curve.(2)Based on the spectral reflectance data provided by this study, the best time to study the spectral reflectance character of rape is the bolting phase of rape.(3) Recognition model of different types of rapes based on canopy spectral was established within550-650nm and670-720nm. Results show as below:the accuracy of model based on Brassica napus is90%, that based on Brassica Rapa is83%, and that based on Brassica juncea is83%; Recognition model of different types of rapes based on leaf spectral was established within530-600nm and660-680nm. Results show as below:the accuracy of model based on Brassica napus is85%, that based on Brassica rapa is81%, and that based on Brassica juncea is80%; Recognition model of different varieties of rapes based on canopy spectral was established within530-670nm. Results show as below:the accuracy of model based on Zhongshuang No.9is79%, that based on Huyou No.15is77%and that based on Xiangyou No.15is78%. Recognition model of different varieties of rapes based on leaf spectral was established within530-600nm and680-720nm. Results show as below:the accuracy of model based on Huyou No.15towards MH15is75%, and that based on Xiangyyou No.15towards MX15is73%.(4) Characteristic wavelength index method can preferably separate different types of rapes. Brassica napus, Brassica rapa, Brassica juncea can be separated pretty good based on canopy spectral and leaf spectral; Then, several varieties of Brassica napus can be distinguished based on canopy level alone; The spectral differences between most of genetically modified rapes and their parents can be identified based on leaf level.