SASP: Segment any strawberry plant, an end-to-end strawberry canopy volume estimation

Authors: HUANG, ZIJING - University Of Florida; LEE, WONG SUK - University Of Florida; ZHANG, PEIGENG - University Of Florida; Jeon, Hongyoung; Zhu, Heping

Submitted to: Smart Agricultural Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/14/2025
Publication Date: 5/15/2025
Citation: Huang, Z., Lee, W., Zhang, P., Jeon, H., Zhu, H. 2025. SASP: Segment any strawberry plant, an end-to-end strawberry canopy volume estimation. Smart Agricultural Technology. 11. Article 101017. https://doi.org/10.1016/j.atech.2025.101017.
DOI: https://doi.org/10.1016/j.atech.2025.101017

Interpretive Summary: The volume of a strawberry plant canopy is an important parameter, and it is often used to predict strawberry yield because the canopy volume can reflect the plant’s health and give an early reliable indicator for yield estimation. The circumferences of the strawberry plants are often used to estimate their volumes. However, manual measurements of the strawberry canopy volumes in a field are tedious and labor-intensive tasks in addition to inaccurate measurements due to human errors. Thus, a simple non-destructive way to measure canopy volumes of strawberry plants would make this task simple and beneficial to strawberry growers to predict accurate strawberry yield. In this research, an automatic method for measuring volumes of strawberry plant canopies with color images was developed by incorporating sophisticated image-processing techniques including a volume compensating factor. The algorithm was evaluated for its accuracy of the canopy volume predictions which were within 5% to 25% of the actual volume. The developed workflow could enable a simple and automated method to predict the canopy volumes of strawberry plants in a high throughput manner with a set of color images captured in a strawberry field.

Technical Abstract: This study presents an end-to-end workflow—Segment Any Strawberry Plant (SASP)—for estimating strawberry canopy volume from multi-view images. The approach utilizes several recent advances in computer vision and 3D reconstruction. First, a Planar-based Gaussian Splatting Reconstruction (PGSR) method is employed to generate high-fidelity 3D point clouds of strawberry plants, offering improved geometric consistency compared to standard 3D Gaussian Splatting. Next, the Segment Any 3D Gaussians (SAGA) framework is adapted with fully automated prompts derived from YOLO (You Only Look Once) detection and color-based prompt selection which can eliminate the need for manual user input in the segmentation process. The resulting point clouds of plant canopies are calculated via concave hull to estimate their volumes. A reference box of known volume is included in the scene as a calibration object, mapping computed volumes from the virtual 3D space into real-world measurements. Experimental evaluations show that the proposed method achieves high segmentation quality and offers volume estimates across multiple plant shapes. This end-to-end pipeline addresses both the labor-intensive nature of manual canopy measurements and the computational complexity of large-scale 3D reconstructions, offering a potential for high-throughput phenotyping and yield prediction in future strawberry cultivation studies.