| The nitrogen content of maize leaves is an important indicator of maize growth.It is one of the main factors that directly affect the growth and yield of maize.The traditional maize leaves nitrogen content is usually detected by chemical analysis method,which is complex in operation,poor in real-time performance and easily generates errors if handled incorrectly.In this paper,spectrum and image techniques are used to analyze and detect the nitrogen content in maize leaves,Provided technical support for the laboratory analysis and planting application of nitrogen content in maize leaves,meanwhile,the detection method of water content in maize leaves is studied,which realizes rapid and accurate detection of the nitrogen content in maize leaves.Main research contents of this paper are as follows:(1)The experiment environment is set up.The samples of maize leaves are obtained by planting experiment,the original spectral data is preprocessed by standard normalization combined with SG convolution smoothing method,and the characteristic wavelengths of three key stages of topdressing(jointing stage,bell mouth stage and heading stage)are extracted by continuous projection algorithm,which provide the basis for reducing data training dimension and improving modeling efficiency.(2)Through multiple linear regressions,BP neural network,GA-optimized BP neural network and random forest,the mathematical models of light reflectance and nitrogen content in maize leaves at full band and characteristic wavelengths are established respectively.The accuracy of the characteristic wavelengths extracted in this paper is verified by comparing the average accuracy of the above models at full band and characteristic wavelengths,and get a spectral nitrogen content model suitable for laboratory detection and analysis.(3)Combined with image technology and convolution neural network algorithm,the automatic modeling and prediction method of maize leaves nitrogen content is studied.The relationship model between maize leaf images and nitrogen content is established,and the automatic images preprocessing and training are realized,finally,an image nitrogen content model suitable for planting applications is obtained. |