YILU YANG et al: DETECTION OF PEPPER LEAVES NITROGEN CONTENT BASED ON CHLOROPHYLL

Nitrogen is a vital element in crops, and its management is a key measure to improve their yield and quality. Real-time monitoring of crop nitrogen content is needed to optimize products and crop production. In this paper we use chlorophyll fluorescence source to obtain pepper leaf image for machine vision and digital image processing and establish a greenhouse pepper leaf nitrogen content rapid non-destructive prediction model. Our results show: i) the 25 chlorophyll fluorescence image characteristic parameters of pepper seedling stage, flowering stage and fruiting stage were significantly related to 18 levels of nitrogen content and 0.01 characteristic parameters, ii) using principal component analysis (PCA) we extracted respectively the main components of the three periods for the model of input variables, iii) using genetic algorithm we optimized back-propagation artificial neural network (BPNN), partial least squares (PLS) and multiple linear regression (MLR) provision method to establish the prediction model of hot pepper leaf nitrogen content. Our conclusions include: i) the correlation coefficients of BPNN and MLR all reached 0.94, ii) the prediction set coefficient was over 0.8, which has a good effect, iii) the effect of PLS model and prediction is not better than MLR and BPNN, iv) optimal prediction model for BPNN model, the modeling correlation coefficient of seedling stage, flowering period and fruiting stage were 0.962, 0.978, 0.945, iv) the correlation coefficient of prediction were 0.908, 0.906 and 0.905, v) compared to the linear regression method like MLR and PLS, nonlinear regression method BPNN was more applicable to establish pepper leaf nitrogen content prediction model, vi) it can reach the desired accuracy and provides an important reference for greenhouse crop nutrition regulation.