Research On Hyperspectral Detection Of Apple Quality Based On Integrated Learning

China is a fruit production and consumption power,fruit cultivation area and yield showed steady growth trend,fruit output value after food and vegetables,competitive advantage,good economic returns.The steady development of the fruit industry plays an important role in promoting the adjustment of agricultural structure,the development of rural economy,the promotion of farmers' income,improving the ecological environment and accelerating the construction of new socialist countryside.Nondestructive testing technology has become a hot trend,and the hyperspectral technology has also made it the best detection method of nondestructive testing technology.It combines the advantages of image technology and spectral technology,an image containing the fruit image information also contains its spectral information,the fruit can be inside and outside the quality of rapid detection.In this paper,apple as a research object,the use of hyperspectral imaging technology on the Apple internal quality parameters of non-destructive testing.In this paper,the Golden variety of apple as the detection object,the number of samples of a total of 300(250 training set of samples,50 predictive set of samples)to sugar,hardness and water content for quality testing,the integrated use of metrology,Statistics,machine learning theory and computer science on the Golden apple quality hyperspectral non-destructive testing process of the characteristics of wavelength selection,correlation analysis modeling,and multi-algorithm integrated learning more in-depth study.The research contents and conclusions are as follows:1)Using the single feature extraction algorithm(continuous projection algorithm,no information variable elimination algorithm,positive self-weight weighted average algorithm,genetic algorithm),the hyperspectral characteristic wavelengths of water content,sugar content and hardness of apple were selected.And then based on integrated learning proposed integrated fusion method to integrate multiple algorithms,taking into account the results of each algorithm wavelength selection.(39 wavelengths)of the "golden handsome" apple,and the predicted coefficient is improved from 0.7889 to 0.8752,and the predicted root mean square error is reduced from 0.2656 to 0.1687.The optimal hardness is the best(27 wavelengths),the predicted results were increased from 0.7094 to 0.8463,from 0.46066 to 0.3059,and the best wavelength(140 wavelengths)of the moisture content was higher than that of the full band,and the prediction was improved from 0.6828 To 0.7835,from 1.1%to 0.48%;moisture content quality,in order to achieve higher accuracy,the selected wavelength more,indicating 450-967nm spectral range of the relevant characteristics of the water less moisture;Compared with the single wavelength extraction algorithm,the integrated fusion algorithm has higher accuracy and the final extraction wavelength is smaller,which indicates that the integrated fusion method is better for the hyperspectral characteristic wavelength.2)First,the partial least squares regression model,the error reverse conduction neural network model,the principal component regression model,the multiple linear regression model,and the support vector regression model of the Golden apple sugar content,hardness and moisture content were established.And the optimal model is the least squares regression model.The optimal model is the error reverse conduction neural network model.The optimal model under the water content is the partial least squares regression model.Then,according to the integrated learning,Bagging integrated framework to determine the Golden apple moisture content,sugar content,hardness of the multi-algorithm integrated prediction model.The prediction accuracy is close to the optimal partial least squares regression model,which is better than the other algorithms.The hardness of the multi-algorithm is 0.8463,0.3045,and the prediction accuracy is close to the optimal error The reverse conduction neural network model is superior to the rest of the other algorithm model.The water saturation ratio is 0.7857,0.48%,and the prediction accuracy is close to the optimal partial least squares regression model,which is better than the other algorithms.It is shown that the integrated model under the integrated learning is feasible for hyperspectral detection of apple quality.Under a certain precision condition,the multi-algorithm integrated model simplifies the manual adjustment of the parameters of the steps,the quality of the target can be a good self-adaptation to ensure stability,and ultimately identified as Golden apple quality hyperspectral detection model.