Estimation Of Phosphorus Content In Rice Leaves Based On RGB Image Processing And Hyperspectral Technology

In order to explore a real-time,rapid and automatic estimation method of phosphorus content in rice leaves,this study used RGB image processing and hyperspectral technology and theory to reveal the relationship between phosphorus content in rice leaves and leaf characteristics and spectral reflectance,and screened leaf characteristic parameters and sensitive bands.Using mathematical statistical analysis and modeling of machine learning methods,build based on leaf color,shape and texture characteristics of rice leaf phosphorus content in qualitative estimation model and based on the leaf spectral reflectance of rice leaf phosphorus content quantitative estimation model,for rice P nutrition of real-time monitoring and further rice field management,provide key technical support and theoretical basis for efficient phosphate fertilization.The main conclusions are as follows:（1）The changes of rice leaf characteristics under different p levels were revealed.In terms of color characteristics,red light,green light,saturation and brightness of rice leaves decreased with the increase of leaf phosphorus content,while hue value increased with the increase of leaf phosphorus content.The eccentricity of shape characteristics decreased with the increase of phosphorus content in leaves.The contrast of texture features decreased with the increase of leaf phosphorus content.Leaf characteristics did not change significantly in rice with excessive P uptake.（2）The spectral response characteristics of rice leaves with different p contents were clarified.In the band range of 780-1300 nm,the spectral reflectance of rice leaves increased with the increase of P content,and the correlation between them reached a very significant level（p<0.01）.It was determined that the near-infrared band was more suitable for predicting P content in rice leaves than the visible band.（3）The best qualitative estimation model of phosphorus content in rice leaves was established.Support vector machine model has the best performance among the four machine learning models used in this study,and its accuracy,recall rate and F1 score are the highest among all models,which are 0.906,0.889 and 0.897,respectively.The estimation accuracy of extreme phosphorus deficiency,moderate phosphorus deficiency and adequate phosphorus levels were 93%,76%and 97%,respectively.（4）The optimal quantitative estimation model of phosphorus content in rice leaves was established.Among the nine spectral indices constructed by band optimization,the difference spectral index DSI(λ₂₂₇₁,λ₉₉₆)based on first-order differential spectrum has the highest prediction accuracy,with a determination coefficient of 0.59.Among the six kinds of machine learning quantitative estimation models,the nonlinear model performs better than the linear model,and the random forest model based on continuous wavelet transform spectrum has a determination coefficient of 0.76,root mean square error and relative analysis error of 0.49 and 2.02,respectively,and the minimalism of the model is the best.