Study On Detection Method Of Rapeseed Maturity Based On Hyperspectral Technology

Rapeseed is one of the important oil crops in China.Combine harvesters are often used for one-time harvesting in production.However,because rape is an infinite inflorescence crop,there will be some immature seeds such as green seeds and yellow seeds during onetime harvesting.In the subsequent drying,it will be transformed into dark or brown seeds,but its chlorophyll content is high,and the oil content is quite different from that of mature rapeseed,which will affect the quality of rapeseed oil.The national standard "GB/T111762-2006" specifies different grades in rapeseed,among which the immature grain is one of the important quality indicators for grading.However,because the mature and immature grains are similar in color after drying and storage,there is a lack of fast and effective in actual production.The detection method is difficult to rate.In this project,we studied the detection method of rapeseed maturity based on hyperspectral information,aiming at the problem that it is difficult to judge their maturity from the appearance after drying and natural placement.The main research contents are as follows:(1)Rapeseed samples preparation and hyperspectral information acquisition.Taking "Huayouza 62" rapeseed as the object,three maturity pods of green ripening,yellow ripening and full ripening were collected.The test samples were obtained by drying according to the storage requirements.The oil content,protein,moisture and other indicators of different mature rapeseeds were measured.These indicators showed that the oil content of different mature rapeseeds was very different.single-grain hyperspectral information collection boards for rapeseed were designed and Compared the effects of different color collection plates on the spectral acquisition of rapeseed.The blue collection board was selected.The hyperspectral spectra of 1500 rapeseeds of different maturity were collected by a hyperspectral camera.The 440 nm band grayscale plot was used as a split band.The OTSU algorithm was used to determine the segmentation threshold,and the rapeseed ROI was obtained,and the spectral data and morphological features of the rapeseed were extracted through the mask.The extraction of texture features of rapeseed is done using GLCM.(2)Analysis and research on the discrimination model of rapeseed maturity by spectral preprocessing and modeling methods.The models of RS,KS,and SPXY were compared,and KS was chosen as the sample division algorithm.The whole band of rapeseed were preprocessed by SG smoothing,first-order derivative,second-order derivative,quasinormal variable transformation,and Detrend individually and in combination.The preprocessed spectrum was combined with ELM,KNN,SVM,PLS-DA,and RF classification algorithms to establish a rapeseed maturity discrimination model.The modeling results showed that the discrimination of rapeseed maturity based on spectral features is feasible.Different preprocessing and model combinations can optimize the modeling effect.Among the established models,D2st-SVM had the best modeling effect,and its training set and test set accuracy rate reaches 98.49%,97.87%.(3)Research on the method of rapeseed maturity discrimination based on characteristic wavelengths.For the redundancy of hyperspectral information and long computing time,the effects of using a total of five feature spectral extraction algorithms,namely,competitive adapative reweighted sampling(CARS),successive projections algorithm(SPA),and interval variable iterative space shrinkage approach(IVISSA)and their combinations,on the dimensionality reduction ability of rapeseed spectra and rapeseed maturity discrimination model after different preprocessing were compared and analyzed.Among the feature wavelength rapeseed maturity discrimination models based on different pretreatments,the number of feature wavelengths selected by IVISSA-SPA based on second-order derivatives was smaller,which was 23.72% of the total number of original rapeseed spectra.And the accuracy of the model was as high as 97.86%.(4)The research of rapeseed maturity discrimination method based on the fusion of hyperspectral image information and characteristic wavelength information.Morphological features of rapeseed maturity discrimination model was established,and the accuracy of the model for rapeseed discrimination was all lower than 70%,and the model was poor.A texture feature classification algorithm based on full-band and feature band was established,and the results showed that the full-band texture feature was better than the feature band,and the combination of four textures was better than the single texture feature discrimination,and the accuracy of its training set and test set reached 93.33% and 80.27%.A rapeseed maturity classification model with fusion of spectral features,morphological features and texture features was established to explore the effect of information fusion on rapeseed maturity classification,and the results showed that the fusion model of spectral and morphological features was better than the separate modeling of spectral and morphological features,and its model accuracy was 98.93%.(5)Matlab APP Designer-based software for rapeseed maturity discrimination was developed.The discriminatory software was developed from the needs of discriminating rapeseed at different maturity levels.This software solves the problem of difficult processing of rapeseed hyperspectral data.The user uploads the hyperspectral image of rapeseed and selects the pre-processing algorithm and feature wavelength extraction algorithm.Based on the uploaded hyperspectral images and the selected algorithm,the software obtains the rapeseed spectra and processes the spectral data,and discriminates the maturity of rapeseed based on the existing rapeseed spectral data and maturity model.