Research On Monitoring Method Of Water Content Of Strawberry Leaves Based On Visual Features

The leaves are an important part of plants,the change of water content can affect the color and geometry of leaves.In this paper,the relationship between the water content of strawberry leaves and the two-dimensional plane features and three-dimensional geometric features of leaf images was studied,and the prediction models of water content,image features and geometric parameters was established,so as to realize the non-contact measurement of water content of strawberry leaves by camera.The main contents of this paper are as follows:(1)32 kinds of strawberry leaf color features and leaf projected area under different water treatment were extracted,and the relationship between these 32 kinds of color features and leaf projected areas and water content of leaves was analyzed by linear regression method,in order to select the best color features that can reflect the change of leaf water content and establish the optimal prediction models of water content of strawberry leaves.The results showed that the relationship between color feature(G-R)/(B-G)and water content of leaves was the most significant,and the correlation coefficient was 0.817 6.The results showed that for each combination of color features and leaf area S*,the correlation coefficient with water content was significantly improved,and the modified coefficient of determination of the combination of variable S* and variable G-R was the highest,which was 0.905 9.The univariate and bivariate prediction models of water content of strawberry leaves were established by the combination of(G-R)/(B-G),S*,S* and G-R.Through the model test,it was found that the coefficient of determination of the univariate prediction model based on(G-R)/(B-G)and S* is 0.846 0 and 0.945 0 in the validation set,while the coefficient of determination of the bivariate prediction model based on the combination of S* and G-R is 0.945 7.Therefore,the bivariate prediction model is better than the univariate prediction model.(2)The 3D geometric characteristics of Strawberry Leaves under different water treatments were measured,and the prediction models of leaf geometric parameters and water content of leaves were established.In order to measure the geometric parameters of leaves,binocular active structured light technology was used to obtain the depth map of leaf image,which was transformed into 3D point cloud and reconstructed the 3D appearance of strawberry.Then the strawberry leaves were segmented by 3D instance segmentation technology,and the plane fitting and spherical fitting of the leaves were carried out.In order to reflect the morphological changes of strawberry in water shortage,the leaf inclination angle and sphere radius obtained by plane fitting and spherical fitting were used to represent the inclination and curl degree of the whole leaves.The relationships between single variable leaf inclination,cosine values of leaf inclination and sphere radius and water content of strawberry leaves were analyzed on the modeling set.In univariate regression analysis,there were significant correlations between leaf inclination,cosine values of leaf inclination and sphere radius and water content of strawberry leaves,and the correlation coefficients were 0.842 9,0.854 6 and 0.880 8 respectively,the standard errors of correlation coefficients were less than 10%.Therefore,the Univariate prediction model of strawberry leaf water content was established based on these three geometric parameters.On the modeling set,the relationship between leaf inclination and sphere radius,and the relationship between cosine values of leaf inclination and sphere radius are analyzed.The results showed that the correction coefficients were 0.914 3 and 0.912 9,both of which were significant.Therefore,the bivariate prediction models of strawberry leaf water content were established based on leaf inclination and sphere radius,and the cosine values of leaf inclination and sphere radius respectively.Through the model test,it was found that the bivariate prediction model had better prediction effect on the validation set.Comparing the two-dimensional and three-dimensional models,it is found that the bivariate model based on 3D geometric parameters of leaf inclination and sphere radius can better reflect the water content of strawberry leaves than other models,so it is feasible to monitor the water content of strawberry leaves non-contact by using visual features.