Detection On Chlorophyll And Nitrogen Contents Of Maize Leaves Using Image And Spectrum Technologies

Maize,as one of the three largest grain crops in the world,directly affects the world’s grain production.Chlorophyll in maize leaves and canopy is the basis of photosynthesis and dry matter accumulation.It can also indirectly reflect nutritional status as an indicator of maize growth and development.Nitrogen is not only an important component of chlorophyll in maize,but also a basic element of protein,nucleic acid and other substances.It is called life element.Therefore,the rapid detection of chlorophyll and nitrogen content in maize leaves is of great significance for guiding maize fertilization and achieving good quality and high yield.With the development of digital image technology and spectroscopic technology,some researchers have applied them to the nutritional diagnosis of crops.However,the existing research about chlorophyll and nitrogen content detection based on digital image technology is limited to one or few varieties,and to less objects.In terms of spectroscopic technology,the application of hyperspectral is much more common.In order to enrich the research about chlorophyll and nitrogen content in maize leaves and provide the basis for the development of portable chlorophyll and nitrogen content detector,the quantitative prediction methods of chlorophyll and nitrogen content in maize leaves were studied in this paper by using digital image and visible/near infrared spectroscopic technology.The main contents and conclusions are as follows:(1)The images were acquired and processed,then the color feature parameters were extracted and transformed.The standard background plate with calibration color blocks was made,and the image data was collected in the designed dark box using the mobile phone camera.The images were smoothed via the median filtering method.The color correction was applied based on the calibrated color blocks group.And then the image of the maize leaf area was separated by using the K-means clustering algorithm to remove the background and the main vein.The color space was converted to extract 9 basic color feature parameters,R,G,B,L,a,B,H,S and V under three color spaces.These parameters were transformed and processed,and25 color feature parameters were finally obtained.(2)The chlorophyll and nitrogen content prediction models were established based on color feature parameters.The correlations between color feature parameters and chlorophyll content and nitrogen content were analyzed.In the detection of chlorophyll content,the best color feature parameter was G-B.Based on this parameter,a linear regression model was established to predict the content of chlorophyll.The mean square root error was 0.3830mg/g.The absolute errors were-0.6162mg/g～0.6432mg/g,and the absolute error was-0.0688mg/g.For the detection of nitrogen content,the best color feature parameter was H.Based on this parameter,a linear regression model for predicting nitrogen content was established.The absolute errors were-0.53%～0.55%,and the mean absolute error was 0.05%.(3)A chlorophyll and nitrogen content prediction model was established based on the selected characteristic wavelengths.The spectral data of the interception range were500nm～900nm,which contained 2095 wavelength points.The S-G smoothing algorithm was used to pretreat the spectrum.The samples were divided by the SPXY method.UVE and SPA algorithms were used to select the characteristic wavelengths respectively.Then the models for predicting the content of chlorophyll and nitrogen were established.In the predicting of chlorophyll content,the best prediction model was UVE-PLS model,and the RPD value was4.2072.Excepting for 1 point’s absolute prediction error was 0.4387mg/g,the absolute errors of the other 30 points were between-0.2537mg/g and 0.1982mg/g,and the mean absolute error was 0.0072mg/g.The higher prediction accuracy was achieved.The best prediction model of nitrogen content was SPA-PLS model,and its RPD value was 2.5453.The prediction absolute errors were-0.47%～0.31%,and the mean absolute error was-0.07%.The better prediction effect was achieved.(4)A model for predicting chlorophyll and nitrogen content based on spectral reflectance under a single band was established.The correlations between spectral reflectance and chlorophyll content and nitrogen content in each single band were analyzed.It was found that the best correlation band with chlorophyll content was 717 nm.Based on the spectral reflectance under this band,a linear regression model was established to predict the content of chlorophyll.The absolute errors were-0.5739mg/g～0.6180mg/g,and the mean absolute error was-0.0091mg/g.The best correlation band of nitrogen content was 549 nm.The result of model verification was that the absolute errors were-0.77%～0.59%,and the mean absolute error was 0.01%.