With the increasing global demand for food, breeding soybean varieties resistant to dense planting is crucial for achieving high and stable yields. Traditional phenotyping methods are limited by insufficient temporal resolution and challenges in dynamic modeling continuity, making it difficult to elucidate the intrinsic relationship between canopy development rate and yield stability. Moreover, existing machine learning models often neglect temporal dependencies in time series predictions, leading to insufficient biological interpretability. This study proposes an innovative approach integrating spatiotemporal deep learning and dynamic modeling to quantify the dynamic changes in canopy parameters using UAV high-throughput phenotyping technology, revealing the key regulatory mechanisms of traits associated with resistance to dense planting. Based on a two-year field experiment (2022–2023) in northeast China (Qiqihaer, black soil region), this study set high (50w plants/ha) and low density (30w plants/ha) treatments across 208 soybean varieties, combined with multispectral UAV imagery (15–18 times per season) and ground-truth data, to develop a time series prediction model for leaf area index (LAI). Comparing the performance of spatiotemporal residual networks (ST-ResNet), long short-term memory networks (LSTM), and traditional random forests (RF), the ST-ResNet model demonstrated significantly superior prediction accuracy (R² = 0.90, RMSE = 0.23 m²/m²), effectively capturing the continuous dynamics of canopy growth through its spatiotemporal feature fusion ability. By fitting the time series curves of LAI, canopy cover (CC), and plant height (PH) with P-spline, 15 intermediate traits (e.g., ΔMean_LAI-mid) were extracted. Mixed models and SHAP interpretability analysis showed that ΔMean_LAI-mid was most correlated with the dense planting yield index (ΔYield, r = 0.51). Furthermore, the high-frequency data acquisition and automated analysis framework using UAVs enabled high-throughput phenotypic screening for 208 varieties per year, significantly improving efficiency compared to traditional methods that rely on manual sampling. This study pioneers the integration of spatiotemporal deep learning with dynamic trait modeling, markedly improving the temporal continuity and stability of LAI estimation compared to traditional single-time-point prediction methods. This advancement allows for more precise quantification of canopy development rates across various growth stages, enabling a systematic analysis of how these dynamic patterns influence resistance to dense planting. By elucidating the dynamic relationship between intermediate traits and yield, this approach offers a high-precision, interpretable phenotypic analysis framework for effectively screening soybean varieties resilient to dense planting.

Winter is a critical period for greenhouse crops production. Multi-source sensors were utilized to quantitatively describe the microclimate distribution patterns and the interactions between crops and greenhouse environment during the winter season. By integrating 3D simulation model, we calculated the energy balance of solar greenhouses, thus assessing the suitability and economic efficiency of several common greenhouse heating devices. The results revealed significant discrepancy in energy demand across different planting areas within the greenhouse, especially in the north-south direction. Significant interactions between crops and greenhouse environment were observed. The presence of cucumber canopies significantly increased relative humidity by 3.0% to 3.8% and negatively impacted air temperature by 1.1 ℃ to 2.5 ℃. Considering the energy balance, initial investment in heating devices, usage period, and utilization effectiveness, the light tube radiators were identified as a suitable heating option. Compared to the unheated greenhouse, the heated greenhouse raised minimum temperatures by 5.2 ℃ to 10.2 ℃, reduced maximum relative humidity by 4.2% to 14.6%, and had a return-on-investment period of approximately 4.7 years. These findings offer valuable insights for multi-source sensor deployment, distributed control, structure optimization, and heating design in solar greenhouse.

The unmanned aerial vehicle (UAV) platform has emerged as a powerful tool in soybean (Glycine max (L.) Merr.) breeding phenotype research due to its high throughput and adaptability. However, previous studies have predominantly relied on statistical features like vegetation indices and textures, overlooking the crucial structural information embedded in the data. Feature fusion has often been confined to a one-dimensional exponential form, which can decouple spatial and spectral information and neglect their interactions at the data level. In this study, we leverage our team's cross-circling oblique (CCO) route photography and Structure-from-Motion with Multi-View Stereo (SfM-MVS) techniques to reconstruct the three-dimensional (3D) structure of soybean canopies. Newly point cloud deep learning models SoyNet and SoyNet-Res were further created with two novel data-level fusion that integrate spatial structure and color information. Our results reveal that incorporating RGB color and vegetation index (Ⅵ) spectral information with spatial structure information, leads to a significant reduction in root mean square error (RMSE) for yield estimation (22.55 ​kg ​ha⁻¹) and an improvement in F1-score for five-class lodging discrimination (0.06) at S7 growth stage. The SoyNet-Res model employing multi-task learning exhibits better accuracy in both yield estimation (RMSE: 349.45 ​kg ​ha⁻¹) when compared to the H2O-AutoML. Furthermore, our findings indicate that multi-task deep learning outperforms single-task learning in lodging discrimination, achieving an accuracy top-2 of 0.87 and accuracy top-3 of 0.97 for five-class. In conclusion, the point cloud deep learning method exhibits tremendous potential in learning multi-phenotype tasks, laying the foundation for optimizing soybean breeding programs.

In quantitative genomic analysis of wheat plant height (PH), the average height of a few representative plants is typically used to represent the PH of the entire plot, which overlooks the variation in height among other plants. Extracting different height quantiles from canopy point clouds can address this limitation. For this purpose, low-cost UAV cross-circling oblique (CCO) imaging, combined with structure-from-motion (SfM) and multi-view stereopsis (MVS), was employed to generate precise canopy point clouds for 262 F5 recombinant inbred lines (Zhongmai 578 ​× ​Jimai 22) across seven environments. Multi-level 3D-PH measurements were extracted from six height quantiles, revealing a strong correlation (mean r ​= ​0.95) between 3D-PH and field-measured PH (FM-PH) across environments. The 90 ​% and 92 ​% height quantiles showed the closest agreement with FM-PH compared to other quantiles. Eleven stable quantitative trait loci (QTLs) associated with multi-level 3D-PH were identified using a 50K single nucleotide polymorphism array. Among these, QPhzj.caas-3A.2 (detected by 3D-PH) and QPhzj.caas-7A.1 (detected by both FM-PH and 3D-PH) represented potential novel loci. KASP markers for these QTLs were developed and validated. Furthermore, within the intervals of QPhzj.caas-5A and QPhzj.caas-3B (both were detected by 3D-PH), two candidate genes associated with PH regulation were identified: TaGL3-5A and Rht5, respectively. Corresponding KASP markers for these genes were also developed and validated. This study highlighted the advantages of 3D model and multi-level 3D-PH in elucidating the genetic basis of crop height, and provided a precise and objective basis for advancing wheat breeding programs.

Quantifying canopy light interception provides insight into the effects of plant spacing, canopy structure, and leaf orientation on radiation distribution. This is essential for increasing crop yield and improving product quality. Canopy light interception can be quantified using 3-dimensional (3D) plant models and optical simulations. However, virtual 3D canopy models (VCMs) have often been used to quantify canopy light interception because realistic 3D canopy models (RCMs) are difficult to obtain in the field. This study aims to compare the differences in light interception between VCMs and RCM. A realistic 3D maize canopy model (RCM) was reconstructed over a large area of the field using an advanced unmanned aerial vehicle cross-circling oblique (CCO) route and the structure from motion-multi-view stereo method. Three types of VCMs (VCM-1, VCM-4, and VCM-8) were then created by replicating 1, 4, and 8 individual realistic plants constructed by CCO in the center of the corresponding RCM. The daily light interception per unit area (DLI), as computed for the 3 VCMs, exhibited marked deviation from the RCM, as evinced by the relative root mean square error (rRMSE) values of 20.22%, 17.38%, and 15.48%, respectively. Although this difference decreased as the number of plants used to replicate the virtual canopy increased, rRMSE of DLI for VCM-8 and RCM still reached 15.48%. It was also found that the difference in light interception between RCMs and VCMs was substantially smaller in the early stage (48 days after sowing [DAS]) than in the late stage (70 DAS). This study highlights the importance of using RCM when calculating light interception in the field, especially in the later growth stages of plants.