Hyperspectrum Detection For Nitrogencontent Of Apple Leaves

The nitrogen content of leaves is an important indicator for estimating growth status of apple tree. Although traditional method for detecting nitrogen content has a high accuracy, its characteristics of poor real-time performance, complex and costly exercise prevent it from a wide range of applications. In contrast, the hyperspectral technology with the characteristics of fast, environment friendly and real-time is becoming an important way to get nutritive elements and detecting growth status of crops. Frist, we measured spectral reflectance for apple leaves using ASD Field Spec 3 spectrometer and nitrogen content of apple leaves using traditional methods in lab. Second, we transformed the original spectrum reflectance of apple leaves to various forms(including the first derivative, reciprocal, logarithm, the first derivative of reciprocal, the first derivative of logarithm) and found five sensitive wavebands. On basis of multiple regression analysis, we obtained four more sensitive wavebands. Again, we divided samples into several groups using cluster analysis in SPSS software and choose the group that had a better correlation with nitrogen content. Finally, we used multiple regression analysis and BP artificial neural network algorithm respectively to construct the retrieval model for nitrogen content of apple leaves. The results showed that the network structure of BP artificial neural network algorithm of 4-10-1contructed the best model. The R2 of training samples is 0.8878 and RMSE is 0.0142.The R2 of testing samples is 0.9001 and RMSE is 0.0150. Therefore, the BP artificial neural network model had a higher accuracy. The results showed that it is an effective way to improve ability of predicting apple tree nitrogen content based on BP artificial neural network algorithm.