Hao Liang

Advance forecasts of minimum greenhouse air temperature can help growers plan cold protection for winter strawberries. Regional weather forecasts describe future atmospheric conditions but do not directly represent greenhouse microclimates, whereas local observations capture historical conditions but lack future weather information. We developed a dual-branch model to combine these complementary sources. One branch processes the preceding 12 h of observations using convolution, bidirectional long short-term memory, and attention; the other extracts forecast features through cascaded convolutions. Using winter strawberry greenhouse data from Lin’an, China, we evaluated minimum temperature predictions over 1–12 h windows. Adding the forecast branch reduced root meansquare error and mean absolute error by 30.38% and 21.12%, respectively, relative to the historical-only model. After hyperparameter selection using the dung beetle optimizer, the final model achieved a coefficient of determination of 0.922, a root mean square error of 1.512 ◦C, and a mean absolute error of 1.041 ◦C for the 8 h window. In an independent greenhouse used for cross-greenhouse evaluation, the corresponding values were 0.885, 1.812 ◦C, and 1.141 ◦C. Combining regional forecasts with local observations improved minimum temperature prediction and could support cold-protection planning in strawberry greenhouses.

Apple seedlings are of great economic importance, yet their susceptibility to water stress poses a serious threat to growth and yield, impacting horticultural productivity. Early detection can help mitigate losses, yet existing plant water stress detection methods are often invasive, time‐consuming, and difficult to apply to the sensitive seedling stage. This study pursued the goal of early, accurate, and non‐invasive identification of water stress (normal, drought, and overwatering) in apple seedling leaves by integrating several analytical techniques: endmember extraction, two‐dimensional correlation analysis, pseudo‐image analysis, and full‐channel hyperspectral imaging data. Using M9T337 apple seedlings as the research object, a total of 570 hyperspectral image data cubes were acquired for analysis and modeling. First, the N‐FINDR algorithm extracted pure spectral endmembers. Second, two‐dimensional correlation analysis revealed the temporal patterns of physiological indicators under water stress. Next, pseudo‐color images from RGB wavelengths were classified by an improved SimAM‐ECA‐MobileNet model and other transfer learning architectures. For the final step of processing whole‐channel hyperspectral data, we employed 1D versions of these same classifiers—specifically adapted for the task, including a 1D SimAM‐ECA‐MobileNet network. Our study proposes a dual‐attention deep learning model based on the MobileNet architecture for accurate classification of plant health status. The network combines the SimAM and ECA attention mechanisms, using a weighted fusion strategy to effectively improve the model's ability to identify plant drought, normal, and overwatered states. This method achieved a classification accuracy of 93.33% on the validation set of pseudoimages, significantly outperforming the baseline MobileNet model (81.67%). The proposed dual‐attention model establishes a robust and efficient framework for early water stress diagnosis in apple seedlings.

Rapid and accurate determination of soil nitrogen (N) content is a critical component of precision fertilization and sustainable agricultural production. Visible-near infrared spectroscopy (Vis-NIR) is widely applied in soil nutrient detection due to its advantages of being rapid, non-destructive, and low-cost; however, its prediction accuracy is often compromised by spatial heterogeneity during cross-regional modeling. To address this challenge, this study proposes a GLI-Fusion dual-branch framework. This framework utilizes the sampling point Geographical Location Identifier (GLI) as a low-cost spatial auxiliary feature and achieves deep fusion of spectral and spatial information through an embedding layer and a cross-attention mechanism, thereby significantly improving the accuracy of NIR-based soil nitrogen prediction.To validate the model's performance, a small-scale soil nitrogen dataset was first constructed via field sampling and laboratory analysis in the Nanhu District, Jiaxing, Zhejiang Province, followed by cross-regional validation using the large-scale pan-European LUCAS 2015 public dataset. The results demonstrated that on the self-constructed Jiaxing dataset, the coefficient of determination (R²) of the GLI-Fusion model on the test set reached 0.8196, representing a 29.85% improvement over spectral-only models. On the LUCAS dataset, the R² further increased to 0.8924, outperforming the best baseline model by approximately 4.3%. Statistical tests confirmed that GLI-Fusion performs significantly better than traditional models across different regions and sample sizes. The study indicates that GLI information can effectively compensate for spectral limitations and enhance model generalization ability, while possessing the advantages of being virtually zero-cost, easily accessible, and highly scalable. This method provides a novel approach for constructing high-precision, region-adaptive soil nitrogen prediction models and lays a foundation for rapid field detection and intelligent agricultural management.

To explore robotic peach picking in different modes, this study examined the effects of various peach picking modes on harvesting force and time. A finite element model of peach harvesting structure was established, and harvesting experiment parameters were based on the Box–Behnken design. Harvesting was simulated to collect response time and force data. Subsequently, the optimal harvesting rate under different picking modes was determined. Different picking modes were tested by simulating identical fruit harvesting in the laboratory at the optimal harvesting speed to determine the peak harvesting force and duration. The Bend mode had the lowest picking pressure and the shortest average picking time at 0.7 MPa and 4.2 s, respectively. The Pull and Twist modes had similar pressures and picking times at 1.2 and 1.1 MPa and 5.2 and 5.6 s, respectively. Harvesting in the orchard allowed for harvesting force and duration measurement under different picking modes. The differences in picking pressure and time among the three picking modes increased compared with those of simulated picking, with specific patterns being observed. Picking pressure appeared at P1max, and differences in picking time were prevalent during separation. This study offers valuable insights for future improvements in harvesting modes and efficiency enhancement.

Near-infrared spectroscopy (NIRS) offers significant advantages for the rapid and non-destructive detection of nutrients in livestock manure slurry. However, conventional models based only on spectral features often show limited robustness under cross-seasonal and multi-farm conditions due to differences in farm source, treatment stage, and complex spatiotemporal background. To improve the accuracy and applicability of total nitrogen (TN) prediction in dairy farm manure slurry, this study used 747 samples collected from 36 large-scale dairy farms in Tianjin, China, covering 24 treatment stages and four seasons, together with sample-contextual information such as farm name, longitude, latitude, and season. Competitive adaptive reweighted sampling (CARS) was applied to select key wavelengths from near-infrared spectra. On this basis, a multi-branch gated fusion deep learning model, MultiScaleSE-GatedCNN, was developed to integrate spectral and sample-contextual information. The model combines multi-scale one-dimensional convolution for spectral feature extraction, separate encoding branches for numerical and categorical inputs, and a gated fusion unit for adaptive weighting of different information sources. Results showed that partial least squares regression remained a strong baseline under single-source spectral conditions, but the proposed deep learning fusion model achieved superior predictive performance after introducing sample-contextual information. Ablation experiments demonstrated that different combinations of sample-contextual information contributed differently to model performance, and the combination of spectra, farm name, longitude, and season yielded the best results. Under this optimal input combination, MultiScaleSE-GatedCNN achieved a test-set R² of 0.905, an RMSEP of 367.389, and an RPD of 3.242. These results demonstrate that integrating NIRS with sample-contextual information can effectively improve the accuracy and robustness of TN prediction in dairy farm manure slurry.

Vigna umbellate, a typical edible and medicinal crop, is rich in polyphenolic compounds with antioxidant, antibacterial, anti-inflammatory, and lipid-regulating activities. However, traditional methods for polyphenol content detection rely on chemical analysis, which is cumbersome and time-consuming, making it difficult to meet the demands of high-throughput rapid detection. Although hyperspectral imaging technology offers the potential for non-destructive and rapid detection, existing analytical methods are often limited by issues such as high spectral band redundancy, insufficient feature extraction, and inadequate model stability, which constrain prediction accuracy and practical application potential. To address this, this study proposes a multi-scale residual convolutional neural network (MS-RCNN) based on competitive adaptive reweighted sampling (CARS) for feature band selection, combined with near-infrared hyperspectral imaging technology, to construct a rapid and non-destructive prediction model for the polyphenol content of Vigna umbellata. The model employs a parallel multi-scale convolutional module to extract spectral features with different receptive fields, and incorporates residual connections and adaptive pooling mechanisms to enhance feature reuse and robustness. Experiments compared the performance of partial least squares regression (PLSR), least squares support vector machine (LS-SVM), multi-scale convolutional neural network (MS-CNN), and MS-RCNN models. The results indicate that the MS-RCNN model based on CARS screening achieved the best prediction performance, with a coefficient of determination (R²) of 0.9467, a root mean square error of prediction (RMSEP) of 0.0448, and a residual predictive deviation (RPD) of 4.33. Compared with the optimal PLSR and LSSVM models, its R² values were improved by 0.2078 and 0.1119, respectively. In summary, the MS-RCNN model proposed in this study enables rapid, non-destructive, and accurate prediction of polyphenol content in Vigna umbellata, providing an efficient technical approach for quality detection of edible and medicinal crops.