Prediction Model Of Nitrogen Content In Apple Leaves Based On Multi-scale Factors

Nitrogen is an important essential element in apple trees,which affects the physiological and biochemical processes,fruit resistance,fruit yield,quality and storage characters.Leaves are the most sensitive organs in the whole tree to soil mineral nutrition.The nutrient analysis and diagnosis technology of apple leaf slice can be used as the basis for nitrogen nutrition gain of apple trees.The general methods of nitrogen detection of crop leaves are mainly chemical analysis and determination in field sampling room,leaf color card method and spectral data analysis.Although the results are accurate and reliable,But the analysis cost is high,the inspection period is long and the time limit is poor.Compared with traditional chemical analysis methods and spectral data analysis of nitrogen content in crop leaves,digital image processing and machine learning technology have the advantages of low cost,high operability and good timeliness.Therefore,the use of digital image processing and machine learning technology to obtain the nitrogen nutrition status of apple leaves in the process of growth can provide information support for apple orchard information management and guidance of precise fertilization.Apple leaf samples were collected from April to November in 2020(a complete growth cycle).Apple leaf images were obtained by digital camera under natural light conditions,and the color features of leaf images in RGB space were extracted.The nitrogen content of apple leaves was determined by Automatic Kjeldahl nitrogen analyzer,and the correlation between color features of leaf images and nitrogen content of leaves was analyzed.The main research work and innovation are as follows:(1)Color features of apple leaf images with different scales were extracted and optimized in RGB space.In order to avoid the problem that the monochromatic component parameters of the image have weak ability to represent nitrogen,a digital camera was used to collect four different scale apple leaf images at flowering stage,young fruit stage,fruit expansion stage and mature stage in the field environment,To further improve the accuracy of monitoring the nitrogen content of apple leaves in RGB space based on image processing technology,14 monochromatic component color combination parameters such as R + G + B were introduced,a total of 17 image color features.Principal component analysis was used to reduce dimension of 17 color feature data of apple leaf images with different scales,and new irrelevant variables were obtained,The key influencing factors of different scales of apple leaf nitrogen content were extracted as the input data of multi-scale apple leaf nitrogen content prediction model.(2)A prediction model of nitrogen content in apple leaves was established based on multi-scale factors.They are: PCA-SVM,PCA-BP,pca-elm three combination models,the output parameters are different scales of apple leaf nitrogen content,analysis and comparison to select the best prediction model of different scales.The results showed that the prediction accuracy of PCA-SVM model was higher than that of PCA-BP and pca-elm in different scales of flowering stage,young fruit stage,fruit expansion stage and maturity stage.Therefore,this study chose PCA-SVM combination model as the prediction model of apple leaf nitrogen content based on multi-scale factors.The penalty parameter C and the width parameter of kernel function of support vector machine model are adjusted by adaptive genetic algorithm σ to optimize.The optimized prediction model of nitrogen content was used for field experiment verification.