| Camellia oleifera is a unique raw material of edible vegetable oil in China.The development of Camellia oleifera planting industry has an important strategic significance for the development of grain and oil in China.With the increase of oil tea output,there are a lot of problems such as picking,storage,processing and quality inspection.Because the fresh Camellia oleifera fruit just picked contains a lot of water,which is not conducive to the storage and processing of Camellia oleifera seeds,and a large number of accumulated Camellia oleifera fruits will cause the mildew of Camellia oleifera seeds,and the internal mildew of Camellia oleifera seeds cannot be judged by naked eyes.At present,the traditional detection method is used to detect the water content and mildew of Camellia oleifera seeds.This method is not only time-consuming and laborious,but also destructive.In this study,fresh Camellia oleifera fruits were purchased from Guizhou,Sichuan,Yunnan,Fujian and other places.96 samples were selected to detect the water content of Camellia oleifera seeds.The correlation of water content of Camellia oleifera seeds was compared by spectral reflectance,and the sensitive spectrum was found and analyzed by spectral modeling.In this paper,taking Camellia oleifera seeds as the research object,SG smoothing,multiple scattering correction,first-order differential,second-order differential and other methods are used to preprocess the spectrum.The effective sensitive spectral wavelength is extracted by principal component analysis and stepwise regression.The extracted sensitive wavelength and the measured value of water content of Camellia oleifera seeds are modeled and analyzed to find the best prediction model of water content spectrum of Camellia oleifera seeds.In addition,the same method was used to establish the spectral model of tea seed mildew.The preliminary test proved that it is feasible to detect the mildew of tea seed by Hyperspectral method.According to the results of the spectral model,the software of on-line quality detection of Camellia oleifera seeds and the quality classifier of Camellia oleifera seeds were developed.The main results are as follows:(1)The spectral characteristics of the water content of Camellia oleifera seeds were observed.It was found that the spectral absorption of the water content was strong in the blue light and the red orange light.The reflectivity of visible band is as high as 0.45,and that of near-infrared band is as high as 0.58.The results showed that there was a good correlation between the change of spectral characteristics and the physical detection value of water content of Camellia oleifera seeds.The sensitive bands with high correlation were 410-450 nm,610-620,780-880 nm and 940-971 nm.The sensitive wavelength of water content spectrum of Camellia oleifera seeds after different pretreatment was obtained by stepwise regression method,and the spectral model was established by using the sensitive wavelength.(2)In this study,four different preprocessing methods are compared and analyzed,and the corresponding optimal model is obtained.The results showed that the best pretreatment method for water content of Camellia oleifera seeds was multiple scattering correction method,and the best model was PLSR model.Different water content gradients in Camellia oleifera seeds were detected,and the spectral sensitive band was found out.The prediction model was established by partial least square regression,BP neural network,radial basis function neural network and other algorithms.The experimental results show that the correlation coefficient of the correction set and the correlation number of the verification set of the partial least square regression model of the water content of Camellia oleifera seeds pretreated by SG smoothing,first-order differentiation,second-order differentiation and multiple scattering correction The mean square errors of correction set and verification set are 0.47,0.34,0.17,0.22 and 0.54,0.35,0.49 and 0.27,respectively.The correlation coefficients of the hyperspectral BP neural network prediction models of water content of Camellia oleifera seeds pretreated by different methods were 0.8335,0.8267,0.8645,0.8830 and 0.8126,0.9112,0.7627 and 0.6789,respectively.The root mean square errors of the correction set and the verification set were1.54,1.31,0.61,1.84 and 1.51,1.15,1.72 and 1.14,respectively.The correlation coefficients of hyperspectral radial basis function neural network prediction models of water content of Camellia oleifera seeds pretreated by different methods were 0.8355,0.8542,0.7519,0.7376 and 0.8136,0.9124,0.7631 and 0.7368,respectively.The root mean square errors of correction set and verification set were 1.27,0.65,0.61,0.94 and 1.38,1.41,0.77 and 0.76,respectively.The results show that the partial least square regression model is better than the BP neural network model and the radial basis function neural network model.The results show that the hyperspectral method can be used to predict the moisture content of camellia seed,and provide an effective basis for online quality detection of camellia seed.(3)The same modeling method was used to establish the spectral prediction model of tea seed mildew,and the better spectral prediction model of tea seed mildew was obtained.Among them,the optimal spectral model of tea seed mildew is the spectral prediction model after the first-order differential treatment.After research and analysis,the model validation coefficient is 0.9476,and the correction set root mean square error is 0.54.(4)Using USART HMI program software,the human-computer interface of the water content detection classifier of camellia seed is developed,and an online detection classifier of camellia seed quality is designed.The quality grader of Camellia oleifera seeds is tested and controlled by the single chip microcomputer module to realize the automatic detection and grading.Through the prediction model,the basic on-line detection of water content of Camellia oleifera seeds is realized,and the feasibility of hyperspectral detection of water content of Camellia oleifera seeds is preliminarily proved. |
