Accurate rice yield estimation is essential for food security. Two key factors affecting estimation accuracy are the long-term upward trend in yield over time and regional heterogeneity across space. Current studies predominantly employ statistical detrending methods (e.g., moving averages, linear regression) to isolate temporal trends. However, such methods rely on prior assumptions about the time–yield relationship and may introduce systematic bias when these assumptions break down. Meanwhile, the individual contributions of temporal and spatial information, and their interactive effects, have not been systematically evaluated within a unified framework. We selected 112 rice-growing counties across six U.S. states (2000–2021), using vegetation index (Normalized Difference Vegetation Index), meteorological indicators (growing degree days, killing degree days, and cumulative precipitation), and spatiotemporal variables (year, longitude, and latitude). We designed six input configurations to compare conventional detrending against direct temporal variable inclusion, testing across four model architectures (Long Short-Term Memory, Random Forest, XGBoost, and Transformer). Results showed that: (1) directly inputting year significantly outperformed detrending across all models, with the combined spatiotemporal configuration achieving the best performance (LSTM R² = 0.61 vs. 0.54 for detrending); (2) year was the most important predictor in SHAP analysis, with spatiotemporal variables ranking higher than most meteorological and remote sensing variables; (3) spatial information consistently improved accuracy and mitigated systematic bias for extreme yield regions; (4) the combined configuration performed best across different states, years (including extreme climate events), and yield levels, achieving near-end-of-season accuracy at the grain-filling stage (1.5–2 months before harvest). This study demonstrates that integrating raw spatiotemporal data directly into deep learning models is more effective than statistical detrending, offering a simpler and more robust approach for large-scale crop yield estimation.

Individual biochemical indicators are insufficient for comprehensive tomato food flavor quality assessment, necessitating multi-parameter models of the core soluble taste matrix. We hypothesized that age stratification of trained sensory assessors would expose differential biochemical variable importance profiles in flavor quality prediction. Accordingly, this study aimed to: (1) construct and compare multiple regression models linking eight biochemical indicators to sensory scores, (2) identify key quality drivers via feature selection, and (3) examine whether age stratification alters the identified sensory drivers. Eight baseline taste indicators across 62 tomato cultivars were evaluated by 30 age-stratified trained sensory panelists (<40 and ≥40 years), using cross-validation to ensure model robustness against small-sample constraints. Partial least squares regression (PLSR), support vector regression (SVR), random forest (RF), and Boruta were applied. Random forest achieved the best performance (R² = 0.82). In the full panel model, key variables were fructose, total free amino acids, and vitamin C. After age stratification, the under-40 group retained these variables, whereas the ≥40 group replaced vitamin C with soluble solids. Fructose and total free amino acids were consistently robust drivers, while total acidity remained least important. Deploying the RF–Boruta framework within an age-stratified context provides a structured analytical framework for investigating flavor perception from biochemical data. These findings suggest that fructose and total free amino acids represent highly robust candidate indicators for flavor quality prediction, while age-stratified variances suggest the utility of integrating demographic-specific metrics into precision breeding frameworks.

The canopy closure stage is a critical phase of rice (Oryza sativa L.) development that influences canopy structure and final grain yield. Accurate and continuous monitoring of canopy closure dynamics is therefore essential for variety screening and cultivation optimization. This study combines unmanned aerial vehicle (UAV) remote sensing technology with deep learning-based semantic segmentation to establish an efficient framework for quantifying rice canopy closure dynamics. UAV RGB images were acquired for 198 hybrid rice varieties during early growth stages and used to build a canopy segmentation dataset. Three semantic segmentation models, i.e., DeepLabv3+, U-Net, and PSPNet, were systematically evaluated. Results show that DeepLabv3+ performed the best and enabled precise extraction of rice canopy features, obtaining a mean intersection over union (mIoU) of 0.86. Based on the extracted canopy coverage, the Gompertz model was utilized to characterize temporal canopy closure trajectories for all varieties, achieving an average R² of 0.978. Subsequently, five key dynamic indicators were derived, including canopy closure limit value (K), initial growth coefficient (a), growth rate coefficient (b), maximum instantaneous growth rate (MGR), and days to maximum growth rate (Tm). K-means clustering analysis was performed on these indicators to categorize all rice varieties into three clusters, disclosing pronounced differences in early-stage canopy development characteristics. Correlation analysis further demonstrated that canopy closure dynamics were closely associated with grain yield. Overall, while acknowledging the limitations of a single-season and single-site dataset, this study provides a scalable and objective framework for quantifying rice canopy closure dynamics, offering valuable support for variety selection, cultivation optimization, and high-yield rice production.

Automated rearing of silkworms (Bombyx mori) using artificial feed in large-scale factory environments represents a crucial step toward the modernization of sericulture in China. Timely and accurate monitoring of silkworm growth and feeding conditions is critical for optimizing rearing management in factory-based systems. This study presents Silkworm-Smart, an integrated intelligent platform that combines machine vision and deep learning to enable automated monitoring of silkworm rearing. At its core is Silkworm-AI, an instance segmentation model specifically designed to detect and segment silkworms and residual artificial feed. Developed upon the YOLOv8s-seg framework, Silkworm-AI incorporates three enhanced modules—Large Selective Kernel Network (LSKNet), Global-to-Local Spatial Aggregation (GLSA), and Context-Guided Downsampling (CGD)—to improve feature extraction and segmentation accuracy. Silkworm-AI achieved outstanding performance, with mAP@0.5 scores of 0.971 and 0.967 for silkworm detection and segmentation, and 0.787 and 0.798 for residual feed detection and segmentation, respectively—outperforming seven state-of-the-art deep learning models. Based on model outputs, the platform automatically extracts and visualizes key rearing indicators such as silkworm count, average size, uniformity, size increment, survival rate, and residual feed area. Furthermore, a machine learning–based regression model was developed to predict cocoon yield using these image-derived indicators, achieving R² values of 0.713 and 0.835 for the 4th and 5th instars, respectively. The Silkworm-Smart platform facilitates data-driven decision-making for rearing management, feed adjustment, and production planning, and holds significant promise for advancing automation, precision control, and sustainability in intelligent factory-based sericulture.

Seed germination is the first critical transition in the plant life cycle, strongly influencing seedling establishment, crop uniformity, and subsequent productivity. However, the genetic basis of seed germination in tomato remains incompletely understood. Here, we performed a genome-wide association study (GWAS) using 319 diverse tomato accessions to identify new loci and candidate genes associated with seed germination. Five significant variants were detected across the genome. Among them, SNP_59508294 on chromosome 3 was consistently associated with the seed germination index (SGI), and accessions carrying the GG haplotype exhibited a significantly higher SGI than those carrying the TT haplotype. Seven α-expansin (EXPA) genes were located within or near the associated interval. Integration of transcriptomic data from contrasting accessions and expression profiles across tomato tissues showed SlEXPA10 to be the most plausible candidate gene. The InDels in the SlEXPA10 promoter were highly associated with its low expression level in low SGI varieties. CRISPR/Cas9-mediated disruption of SlEXPA10 significantly reduced the rate and speed of seed germination relative to the wild type. Collectively, these results identify SlEXPA10 as a positive regulator of tomato seed germination and provide a candidate locus for improving seed vigor and emergence uniformity.

Mastication quality is a key determinant of citrus eating quality, but conventional sensory evaluation and instrumental texture analysis are subjective, destructive, or time-consuming. This study combined magnetic induction tomography (MIT) with machine learning to assess citrus mastication quality non-destructively. MIT signals were acquired from 300 ‘Miyagawa’ satsuma mandarin fruits at 21 frequencies (15–35 MHz). Puncture force, shear force, and chewiness were measured instrumentally. These measurements and sensory scores were integrated into an Analytic Hierarchy Process (AHP)-weighted comprehensive mastication score (0–10), which was also divided into four quality grades (A–D). Feature selection ranked raw features by univariate linear-regression F-test p-values for the regression targets or absolute Kendall’s τ values for the classification target. The top 5% of ranked features were retained, and PCA was then fitted to the retained features. Feature distributions were summarized using pooled out-of-fold predictions from nested cross-validation, with five-fold cross-validation used only for inner Bayesian hyperparameter tuning. SVM achieved the highest R² for puncture force (0.825) and shear force (0.762), Random Forest achieved the highest R² for chewiness (0.849) and the highest grade-classification accuracy (64.3%) and macro-averaged AUC (0.871), and XGBoost achieved the highest R² for the comprehensive mastication score (0.835). These results demonstrate the feasibility of combining MIT with machine learning for non-destructive assessment of ‘Miyagawa’ satsuma mandarin citrus mastication quality and identify frequency–spatial patterns that may guide sensor optimization and future grading applications in other citrus fruits.