3D reconstruction and characterization of cotton bolls in situ based on UVA technology

Phenotypic traits at the organ scale hold significant importance in the realm of plant breeding, notably in evaluating genetic diversity, selecting innovative cultivars, and forecasting potential yield. Ground platforms, despite delivering accurate results in organ-scale phenotypic studies, frequently exhibit limitations concerning efficiency and adaptability, and can adversely affect the soil. The Unmanned Aerial Vehicle (UAV) platform emerges as an innovative medium, distinguished by its high-throughput and adaptability, thus offering immense prospects in phenotypic research. Nonetheless, its precision within the organ scale warrants further investigation. As a notable economic plant species, cotton (Gossypium spp.) provides the textile industry with its essential raw material through the fruit it bears, the cotton boll. This study, employing the characterization of cotton boll breeding materials as an example, scrutinizes the capability of UAV platforms in executing precise enumeration and yield estimation of three-dimensional (3D) cotton bolls in agricultural fields. An innovative route, the Cross-circling oblique (CCO) route, was employed in conjunction with lightweight UAVs for reconstructing a 3D point cloud of cotton bolls in the field, and juxtaposed against the traditional nadir route. A comprehensive process was also developed for characterizing the quantity, volume, and spatial distribution of cotton bolls, further facilitating yield estimation. The CCO-produced point cloud demonstrated a superior completeness compared to the nadir-produced point cloud, proving capable of effectively mitigating occlusion in the middle and lower sections, thereby procuring a more comprehensive 3D cotton boll. In terms of cotton boll counting (R2 = 0.92 vs R2 = 0.73) and yield estimation (seed cotton: R2 = 0.7 vs R2 = 0.62, lint cotton: R2 = 0.75 vs R2 = 0.55), the precision of the CCO-derived point cloud surpassed that of the nadir-derived point cloud considerably. Notably, yield estimations were more precise when based on cotton boll count for seed cotton, and on boll volume for lint cotton. Utilizing the high-precision CCO-produced cotton boll point cloud, an analysis was conducted on the spatial distribution of cotton bolls for all breeding materials, and the yield was subsequently projected. The contribution of this study lies in the high-throughput capture of organ-scale traits of field cotton bolls using the UAV-based CCO route, enabling accurate yield estimation. This approach may also be applicable to extracting organ-scale traits of other field crops. It demonstrates the immense potential of combining UAV platforms with the CCO route for high-throughput acquisition of crop organ-scale traits, which could become a powerful tool in plant breeding.