Fruit Moisture Detection Model In Southern Xinjiang Based On Multi-angle Spectral Imaging

Southern Xinjiang fruit due to the long light time,day and night temperature difference and other unique natural and geographical conditions to form a good quality,southern Xinjiang fruit due to the long light time,day and night temperature difference and other unique natural and geographical conditions to form a good quality,the quality of different regional fruits is quite different due to different growth angle,orientation and light conditions.This paper studies the key problem of fruit spectral reflectance difference under different light source angles.Based on multispectral imaging techniques the quality detection test of multi-angle spectral imaging of red jujube,apple,hanging and dried apricot in southern Xinjiang was carried out,the main completed research contents are as follows:First,build a multi-angle spectral fusion quantitative detection and test platform,mark the different quality of fruit by classification;set of multi-angle spectral data to establish the characteristic angle database;multispectral angle fusion reflectance for each fruit was obtained from the image.Preprocessing of the extracted multi-spectral angle-fusion reflectance data,by exploring the polarization characteristics of the three fruits and the change rules of the spectral curves at different angles,the abnormal angles of three fruits are found to improve the modeling accuracy of polarization angle fusion and then the multi-angle polarization feature model is established.The training set correlation coefficients of the jujube modeling results,Rp,R2 and Rmsep,were 0.9779,0.9562,0.0112.The training set correlation coefficient of apricot modeling results,Rp,R2 and Rmsep were 0.9825,0.9665 and 0.0559.The training set correlation coefficients Rp,R2,and Rmsep were 0.9826,0.9656,and 0.0189.The multi-angle polarization model of these three fruits is inverted,multi-angle polarized water inversion diagram of fruit was obtained for quantitative spectral detection.The modeling results of angle fusion data at different angular resolutions show that: by modeling different fruits after angle fusion,modeling accuracy has been greatly improved to achieve what can be used for quantitative detection standards.The relationship between angular resolution and modeling accuracy is not necessarily proportional(Reference the apple and apricot angle fusion(every20)(every 30)in the comparison table).Therefore,the best Angle of different fruits is not exactly the same.Data per 10 fused perform better modeling than every 20,30.It shows that the three kinds of fruit in this experiment can achieve good results at the angular resolution of 10.To verify the feasibility of the laboratory conclusions,Jujube canopy test for verification,five polynomial modeling and comparison between jujube single-angle data and multi-angle fusion data,the results are as follows: The determination coefficient of individual angles(0,45,90,135)R2 is 0.6509,0.72297,0.56247,0.69252 respectively.The data modeling determination coefficient after 4 angles was increased to 0.85837.The contrast determination coefficient R2 per 90 angles and that per 45 angles is increased from 0.73214 to 0.85837.To prove that the influence of different angular resolution of outdoor canopy on modeling accuracy is consistent with indoor conclusions.That is the higher the angular resolution,the higher the modeling accuracy.The water inversion distribution map is obtained from the prediction model,the inversion water varies by 9% between 12% and 28%.The accuracy of the model is consistent with the conclusion of the indoor test.Under different outdoor weather and light conditions,the characteristic angle and polarization direction were selected to further improve the angle resolution and other methods to reduce the error of the outdoor detection model.It has important reference value for the quantitative detection of large area UAV in southern Xinjiang.