In the study titled “Eff-3DPSeg: 3D Organ-Level Plant Shoot Segmentation Using Annotation-Efficient Deep Learning,” researchers have developed Eff-3DPSeg, a deep learning framework for plant organ segmentation. They leveraged point clouds taken from individual plants through a Multi-view Stereo Pheno Platform (MVSP2) and annotated them using a Meshlab-based Plant Annotator (MPA).
Eff-3DPSeg Development and Testing
Eff-3DPSeg uses a two-step process. First, the researchers reconstructed high-resolution point clouds of soybean plants using a low-cost photogrammetry system and developed a Meshlab-based Plant Annotator for plant point cloud annotation. After this, they used weakly supervised deep learning for plant organ segmentation. The researchers trained the model with just approximately 0.5 percent of labeled points, augmented it utilizing Viewpoint Bottleneck loss, and extracted three phenotypic traits: the leaves’ length, width, and stem diameter.
Eff-3DPSeg Performance and Future Directions
Researchers tested the framework’s performance on various growth stages on soybean plants and compared it with fully labeled techniques on tomato and soybean plants. While the stem-leaf segmentation results were accurate, there were small misclassifications at junctions and leaf edges. The approach performed better with larger training sets and in less complex plant structures. However, the study faced limitations of data gaps and the need for separate training for different segmentation tasks, which the researchers plan to refine in the future. They also aim to expand the range of plant classifications and growth phases suited for the framework.
Conclusion: Potential of Eff-3DPSeg
The Eff-3DPSeg framework represents a significant step forward in 3D plant shoot segmentation. With its efficient annotation process and precise segmentation capabilities, it shows great potential for enhancing high throughput and overcoming the challenges of expensive and time-consuming labeling processes through weakly supervised deep learning and innovative annotation techniques.