Precision Agriculture Сельское хозяйство и биологические науки (общие)
Enhancing seeding efficiency using a computer vision system to monitor furrow quality in real-time
Аннотация
Purpose Crop residue interference (hair-pinning) poses a critical barrier to optimal seed furrow formation in conservation tillage, yet currentmanual assessments prevent real-time equipment optimization. The purpose of this study was to develop a high-speed computer vision (CV) methodology to answer the following research question: Can real-time semantic segmentation dynamically quantify seed trench cleanliness and evaluate row cleanerperformance accurately enough to enable autonomous seeding adjustments and eliminate labor-intensive manual assessments? Methods A high-definitionvideo acquisition system was integrated onto an air seeder. The study benchmarked five semantic segmentation architectures (SegFormer, DeepLabV3+,PSPNet, U-Net, and K-Net) to identify a model capable of balancing segmentation accuracy with the strict latency constraints of high-speed planting. Thismethodology was subsequently applied as a proof-of-concept to test four prototype pre-production row cleaner configurations. Results The TransformerbasedSegFormer emerged as the optimal architecture, achieving an 80.9% Soil Intersection over Union (IoU) with an exceptionally fast inference time of11.75 ms. When applied to the row cleaner prototypes, the system revealed that the optimal configuration increased soil exposure to 80.3%, compared to abaseline operation that retained 61.1% residue cover. Conclusion Real-time semantic segmentation via this CV system effectively resolves the complexvisual challenges of furrow analysis. By proving its viability for real-time deployment, this methodology provides a foundational perception layer crucialfor the future of autonomous seeding equipment.