基于影像分割的覆膜玉米叶绿素含量反演OA北大核心CSTPCD

[Objective]In order to quickly and accurately monitor chlorophyll content of film-mulched maize,explore whether the removal of film and shadow background pixels can improve the accuracy of chlorophyll content inversion with spectral and texture features.[Method]This study was based on multi-spectral remote sensing image data of unmanned aerial vehicle(UAV)and took chlorophyll content of film-mulched maize at seedling stage,jointing stage,tasseling stage and filling stage as objects.The support vector machine supervised classification was used to segment image background pixels and maize pixels,analyze the influence of background pixels on the spectra of maize canopy,the vegetation index and texture features of all pixels and maize pixels images were calculated and the better variable input was screened,and the inversion model of leaf chlorophyll content was established by using three machine learning algorithms,partial least squares,support vector machine and BP neural network.[Result](1)Background pixels in the multispectral images at seedling stage,jointing stage,tasseling stage and filling stage had significant effects on the spectra of maize canopy.(2)The inversion accuracy of vegetation index,texture feature and vegetation index + texture feature as variable input based on maize pixels image extraction was better than that of all pixels image(R2 for optimal model was increased by 0.078,RMSE and MAE were decreased by 0.060 and 0.055 mg·g-1,respectively,and R2 for verification was increased by 0.109,RMSE and MAE were reduced by 0.075 and 0.047 mg·g-1,respectively.(3)The modeling accuracy based on maize pixels image with spectral features + texture features as variable inputs was significantly improved over the modeling accuracy using only spectral features or texture features as variable inputs;The BP neural network model with spectral features + texture features as variable inputs had the highest accuracy(R2,RMSE and MAE were 0.690,0.468 mg·g-1 and 0.375 mg·g-1,respectively).[Conclusion]The multispectral image spectral and texture feature data of UAV with removing background pixels and combined with BP neural network can better realize the inversion of chlorophyll content of film-mulched maize.The results can provide theoretical reference for quick and accurate retrieval of leaf chlorophyll content of film-mulched maize by UAV remote sensing.