Monitoring and experimental verification of strawberry leaf moisture content using 3D geometric features

Abstract: Water is indispensable for plant growth, and water shortage will affect the yield, growth, and quality of plants. The rapid and non-destructive detection of water content in plants is of great significance to scientific guidance of irrigation and to improve crop yield. There are many methods to detect and estimate moisture content in plants. The image processing by using the projected area and color space of plant leaves in two-dimensional images is widely applied to detect water content. With the development of 3D point cloud technology, it has become an inevitable trend to use 3D information to study crop growth status. In this study, the monitoring models of strawberry leaf moisture content based on 3D geometric parameters were used to predict the leaf moisture content. The water content of 49 pots of strawberry "Ningyu" with different water treatments was measured. Geometric parameters of strawberry leaves were extracted by using real-time on-line 3D modeling to analyze the relations between water content and these parameters. Firstly, a binocular camera was set to capture the depth maps of strawberry leaves, which were converted into 3D point cloud images. Secondly, the strawberry leaves were segmented by using the 3D-SIS instance segmentation method. Thirdly, the random sampling consensus algorithm plus the global least-squares method was used to fit planes with the segmented 3D leaves to obtain the leaf inclination angle and corresponding cosine value. The blade spherical surface fitting method based on algebra was used to obtain the sphere radius to indicate the bending of leaves due to lack of water. Finally, the relation between water content and geometry parameters was analyzed quantitatively by using univariate and multivariate linear regression. In the univariate linear regression analysis, leaf inclination angle, cosine value, sphere radius, and leaf moisture content were linearly correlated, and the determination coefficients were 0.842 9, 0.854 6, and 0.880 8 respectively. In multiple linear regression analysis, the relation between water content and sphere radius plus leaf angle, and sphere radius plus cosine value was analyzed. The results showed that there was a significant correlation between the spherical radius, the leaf inclination angle, and the water content, and the modified coefficient of determination was 0.914 3. The prediction models were tested on the validation set and the results showed that among the above-said models, the one established by spherical radius and leaf inclination angle was best. The root mean square error was merely 0.015 8, and the determination coefficient was as high as 0.953 4. The proposed method could quickly detect the water content of strawberry leaves, providing an effective method for non-contact measurement of strawberry water content and technical support for the accurate acquisition of agricultural situation information. With the help of 3D geometric parameters of leaves, there was a prospect for the early application of plant water deficit control. The prediction model of leaf water content established by regression should have a certain application value in the field of ecology and crop protection.