| As the main economic crop,brassica napus plays an important role in national economy.In the process of rapeseed growth,chlorophyll content has a direct effect on the photosynthesis of crops,which is an important agronomic parameter to measure the growth of rapeseed,but the actual measurement procedure is complicated and the timeliness is poor.Rapid and non-destructive dynamic monitoring of chlorophyll in rapeseed can be achieved by UAV remote sensing platform,which plays an important guiding role in formula fertilization and nutritional diagnosis of rapeseed.In order to explore a quick and convenient method to estimate chlorophyll content in Brassica napus for efficient crop monitoring,this study focused on the following contents:(1)In order to remove the influence of soil background on color features,the cropped image needs to be segmented to preserve the green part of the image.In this paper,the optimization of K-means clustering segmentation(WSSA-Kmeans)based on the improved sparrow search method is proposed,and the OTSU threshold segmentation method combined with EXR,EXG,EXGR and CIVE four vegetation indices and the traditional Kmeans clustering algorithm are compared to evaluate the image segmentation performance of field rape.WSSA-Kmeans has the best performance and the lowest average error rate(3.8%).(2)Construct the feature parameters of the image,and propose the original band combination of this paper after feature selection and analysis.The results showed that the color characteristic parameters B/(R+ g-b),B/(R+G),B,B/G,(g-b)/(G+B),g-b,R/(R+ g-b)and R/(R+ b-g)obtained from the R(red),G(green),B(blue)three channel components were significantly correlated with the measured chlorophyll value SPAD.Among them,the band combination B/(R+G-B)proposed in this paper has the highest correlation coefficient with SPAD value,which is 0.749.(3)Establish the chlorophyll prediction model of field rape.According to the calculation,46 color characteristic parameters were combined.And with the ratio of 7:3,189 samples were used as modeling to construct the chlorophyll content estimation model of rapeseed,and 81 samples were used as tests to verify the accuracy of the model.Univariate regression model,multivariable stepwise regression model,support vector machine model,random forest model,and the improved random forest regression model(SA-RFR)based on adaptive feature weight and hyperparameter optimization proposed in this paper are constructed respectively.Model comparison shows that the results of SA-RFR model are better(R~2=0.811,RMSE=3.03,RE=6.45%).In order to understand the spatial distribution of chlorophyll content in the field more intuitively,the prediction results of the optimized model are inverted into the visible light image of the UAV,and the grade distribution map of chlorophyll content is drawn.This study provides a new method for rapid estimation of chlorophyll content in rapeseed,and provides an auxiliary means and reference for field management to discriminate measured chlorophyll content. |
