Abstract:

Chlorophyll content is an important index of characterization for plant growth. The traditional chlorophyll content determination methods mainly include spectrophotometry and chlorophyll meter method （SPAD-502）. The spectrophotometric method can determine chlorophyll content accurately but time-consuming，laborious and damaged blades；the chlorophyll meter method acquire chlorophyll content rapidly but the measurement area is limited and repeatedly. Digital image processing technology emerges in response to the needs of times，which provides a scientific basis for diagnosis of the apple tree physiology. The apple leaves were collected from 60 new shoots prosperous long-term apple trees in Mengyin of Shandong Province，and on the same night，the image were taken using a digital camera. The chlorophyll contents were measured by the traditional chemical analysis in laboratory，including Chl.a，Chl.b，Chl.(a + b) and the SPAD value. There was a certain degree difference in the correlation analysis of chlorophyll content with color data，in order to increase the accuracy of chlorophyll content estimation for apple trees，the color parameters were combined based on the RGB values extracted from histogram in Photoshop CS6.0（Adobe System，Inc.）. Then the correlation analysis method was used to select the sensitive color parameters（R，red；G，green；B，blue）. Compared to the trichromatic color values，the combination of RGB values considerably improved the correlation coefficients. Among the color parameters，B，B/R，G，B/G，G/（R + G + B），B/（R + G + B），（R–B）/（R + B），（G–B）/（G + B），（R–B）/（R + G + B）and（G–B）/（R + G + B）values were correlated significantly with chlorophyll content respectively. The correlation coefficients（R）of these essential color parameters were almost above 0.325，which suggested that leaf color parameters were significantly correlated with chlorophyll content（P = 0.01）. According to the result of systematic analysis and diagnosis，the univariate regression model and support vector machine（SVM）regression model method acts as a non-parametric regression technique，which analyzes the fitting the degree of relationship between predicted values and measured values using the determination coefficient（R²），root mean square error（RMSE）and relative error（RE）. In the univariate regression model，the fitting coefficients of the chlorophyll content based on the G/（R + G + B）color parameter achieved the highest R² = 0.736，as Chl.(a + b) > Chl.b > Chl.a > SPAD；the R² of other color parameters differed from 0.482 to 0.742，the improvement of the univariate regression model stability were needed. The SVM model has the advantage of minimized parameter setting and model structure risk. The effect of the SVM model was good for the estimation of chlorophyll content of apple trees，including the fitting coefficients（R²）between estimated and measured value were 0.8754，0.8374，0.8671 and 0.8129，respectively，RMSE were 0.0194，0.0350，0.0497 and 0.9281，respectively，RE were 0.8059%，1.7540%，1.1224% and 0.8059%，respectively，which demonstrated that the SVM estimation had a higher accuracy. By the verification analysis，the SVM based on sensitive color parameters for Chl.a had the best estimation，including R² = 0.8275，RMSE = 0.0293 and RE = 1.8529%. The conclusion were drawn that application of digital cameras based on RGB color model can estimate apple leaf chlorophyll content rapidly. The potential of the imaging system with apple leaves has been discussed in the article. The results can provide the technical support for precise management of orchard.

Key words: apple, leaf, color parameter, chlorophyll content, SPAD value, support vector machine （SVM）, digital camera