Detection Of Nitrogen Content In Apple Tree Leaves Based On Digital Image Processing Techniques

Nitrogen can be utilized to strengthen photosynthesis of fruit trees in orchard management,and reasonable utilization of nitrogen is an effective measure to improve quality and yield of fruits.With continuously extension in scales of orchards in our country,urgent improvement is needed,instead of the traditional extensive type orchard fertilization management method.At present,there are various problems in apple cultivation such as applying nitrogenous fertilizer,insufficient fertilization,excessive fertilization and unreasonable fertilization period,leading to deterioration on yield and quality of fruits,high planting cost and serious environment pollution.For this reason,nitrogen nutrition detection and optimized fertilization management play critical roles in modern orchard management.Most researches on nitrogen content detection of fruit tree leaves are conducted based on analysis on photochromic information of leaves,the color characteristics on which are greatly influenced by illumination,showing different reflected light characteristics within different spectral regions,giving rise to influences on analysis accuracy.It is difficult to collect characteristics reflecting nitrogen deficiency degree such as shoot growth situations and leaf shapes for further analysis relying on only methods such as hyper-spectrum and chlorophyll meter.For this reason,it is especially important to research on morphological characteristics of fruit trees with minor influences by collecting environments,which are as the characteristic parameters for detection of nitrogen content of fruit trees,so as to explore a leaf nitrogen content estimation method with low cost and strong universality.This paper adopts digital image processing technology,selects and analyzes morphological characteristic parameters of leaves on the new shoots and characteristic parameters of leaf node pitches from apple trees with treatment of various nitrogen application levels in visible spectrum,and establishes an estimation model of nitrogen content in apple tree leaves,so as to realize the injury-free rapid detection of nitrogen content in apple trees,determine the nutritional status of apple growth,and thereby provide guidance for orchard operators to fertilize as needed.The main research work and innovations are shown as follows:(1)This paper suggests an extraction method of morphological characteristic parameters of leaves and characteristic parameters of leaf node pitches of the new shoots on apple trees,based on image processing technology.Images of leaves and leaf node pitches of the new shoots are taken and processed by a series of image processing technologies including adaptive filtering noise reduction,color image segmentation,morphological processing,and characteristic extraction,in order to extract the morphological characteristics of leaves and characteristics of leaf node pitches from the images.(2)This paper establishes an estimation model of leaf nitrogen content based on leaf areas of the new shoots.Firstly,correlation analysis and significance test are conducted between morphological characteristics of the collected leaves of the new shoots and the nitrogen content,and get the result that characteristics of leaf areas have a strong correlation with leaf nitrogen content;secondly,a regression model of leaf nitrogen content is established based on leaf areas of the new shoots with linear regression,polynomial regression,and random forest regression;finally,the determination coefficient,root-mean-square error,and residual analysis are adopted to evaluate the precision and reliability of each model.(3)This paper establishes an estimation model of leaf nitrogen content based on leaf node pitches of the new shoots.Firstly,correlation analysis and significance test are conducted between the collected leaf node pitches of the new shoots and nitrogen content of the leaves,and get the result that the correlation between leaf node pitches of the new shoots and leaf nitrogen content is highly significant;secondly,a regression model of leaf nitrogen content is established based on leaf node pitches of the new shoots with linear regression,polynomial regression,and random forest regression;finally,the determination coefficient,root-mean-square error,and residual analysis are adopted to evaluate the precision and reliability of each model.The tests verified that the extraction method of morphological characteristic parameters of leaves and characteristic parameters of leaf node pitches on the new shoot is precise and reliable,which can provide accurate characteristic parameters for the estimation model of the leaf nitrogen content of apple trees.The root-mean-square error RMSE values of leaf area,leaf length,leaf width,and leaf pitch got by the method and measured values are 0.684,0.132,0.104,and 1.143 respectively,and the values of determination coefficient R~2 are 0.983,0.985,0.978,and 0.969.The correlation analysis and significance test of area and pitch characteristics of the leaves on the new shoots from groups of various nitrogen application levels are seen with results of strong correlation.Samples from test sets of each nitrogen level are used for testing regression models,and results show that the random forest regression model established in this paper is seen with favorable fitting effects in the groups of various nitrogen application levels.In the random forest regression model based on leaf area of the new shoots in vigorously growing period of spring,R~2=0.7324 and RMSE=3.7782,and of autumn,R~2=0.7331 and RMSE=3.7849.In the random forest regression model based on leaf pitch of the new shoots in vigorously growing period of spring,R~2=0.8386 and RMSE=3.3769,and of autumn,R~2=0.8335and RMSE=4.0026.Residual tests have also proved that the established random forest regression model is precise and reliable.