Bi Zhao

Nondestructive assessment of finished black tea is challenged by limited reference samples and highly correlated spectra. This study examined whether compact wavelength selection and selective residual modeling could support the prediction of chemically distinct quality constituents. We developed FB-CARS–PER-BHNet, combining frequency-aggregated, block-constrained competitive adaptive reweighted sampling (FB-CARS), a partial least squares regression (PLSR) backbone, and evidence-gated residual learning. Forty-seven finished black tea samples were represented by 243 visible–near-infrared wavelengths (401.05–999.20 nm) and evaluated by outer leave-one-out validation for water extract, epigallocatechin gallate (EGCG), epicatechin gallate (ECG), and lysine (Lys). Held-out R² was 0.9068 for water extract, 0.9051 for EGCG, 0.8644 for ECG, and 0.7553 for Lys. FB-CARS reduced spectral redundancy, while compact PLSR models approached full-spectrum accuracy using 20, 25, 15, and five wavelengths, removing 89.7–97.9% of the original variables. Component-wise ablation showed the largest mean R² decrease when FB-CARS was replaced; PLSR captured the dominant quantitative structure, and evidence gating selectively admitted residual correction. The complete framework achieved higher target-wise R² point estimates than the evaluated standalone regressors (mean R² = 0.8579). Spatial projections of five held-out samples illustrated local PLSR responses without establishing within-sample chemical accuracy. The findings support further evaluation of compact spectral screening, while larger independent datasets are needed before routine quality-control use.

Antimicrobial resistance (AMR) in livestock is a major contributor to the global AMR crisis, yet little is known about its dynamics in high-altitude pastoral systems. We performed deep metagenomic sequencing of 100 fecal samples from Tibetan Awang sheep reared under grazing (aw_fm) and captive (aw_qs) conditions. Microbiome profiling revealed striking community shifts: grazing sheep were enriched in Bacteroidetes and Firmicutes, whereas captive sheep showed expansion of Proteobacteria, particularly Acinetobacter , suggesting dysbiosis. The resistome comprised 302 unique ARGs, dominated by rpoB2 (43.3%), Bifidobacterium _ adolescentis _ rpoB (11.2%), and ugd (10.2%). Grazing sheep carried ARGs mainly against rifamycins and peptide antibiotics, reflecting natural selective pressures, while captive sheep exhibited significantly broader resistance, including tetracyclines, macrolides, and fluoroquinolones ( p < 0.05). Enrichment of efflux pump genes ( MexK , adeJ ) in captive sheep highlighted a shift toward multidrug resistance. These findings demonstrate that rearing practices profoundly restructure the gut resistome, underscoring the need for targeted antibiotic stewardship in high-altitude livestock systems.

Timely detection and regular maintenance of road cracks are critical for road and traffic safety. However, existing detection methods face challenges such as varying target scales, large model parameters, and poor adaptability to complex backgrounds. To address these issues, this study proposes an enhanced GSB-YOLO model. Inspired by the concepts of linear transformation and long-range attention mechanisms, a lightweight network structure was designed to reduce model parameters in the backbone network, thereby improving detection efficiency. Additionally, a novel SMC2f module was introduced in the neck structure, which calculates the “energy” of each neuron in the feature map, evaluates its contribution to the detection task, and dynamically assigns weighted coefficients. This method enhances the model’s detection robustness in complex backgrounds and effectively addresses the issue of insufficient emphasis on positive samples. Furthermore, through the optimization of the Path Aggregation Network (PAN) and the Bidirectional Feature Pyramid Network (BiFPN), efficient multi-scale feature fusion is achieved, further strengthening the model’s capacity to represent crack features at various scales. Experimental results indicate that the proposed GSB-YOLO model improves the mean average precision (mAP) in road crack detection tasks by 3.2%, demonstrating its significant application value in road crack detection and traffic safety assurance.

Oxidative stress-induced photoreceptor cell death is closely associated with the etiology of age-related macular degeneration (AMD), and sodium iodate (SI) has been widely used as an oxidant stimulus in AMD models to induce retinal pigment epithelium (RPE) and photoreceptor cell death. However, the mechanism underlying SI-induced photoreceptor cell death remains controversial and unclear. In this study, we elucidate that ferroptosis is a critical form of cell death induced by SI in photoreceptor-derived 661W cells. SI disrupts system Xc⁻, leading to glutathione (GSH) depletion and triggering lipid peroxidation, thereby promoting ferroptosis in photoreceptor-derived 661W cells. Additionally, SI enhances intracellular Fe²⁺ levels, which further facilitates reactive oxygen species (ROS) accumulation, making the 661W cells more susceptible to ferroptosis. Exogenous GSH, as well as specific inhibitors of ferroptosis such as Fer-1 and antioxidants like NAC, significantly attenuate SI-induced ferroptosis in photoreceptor-derived 661W cells. These findings provide new insights into the mechanisms of ferroptosis as a key pathway in SI-induced photoreceptor-derived 661W cell death.

Timely detection of plant diseases is crucial for agricultural safety, product quality, and environmental protection. However, plant disease detection faces several challenges, including the diversity of plant disease scenarios and complex backgrounds. To address these issues, we propose a plant disease detection model named PYOLO. Firstly, the model enhances feature fusion capabilities by optimizing the PAN structure, introducing a weighted bidirectional feature pyramid network (BiFPN), and repeatedly fusing top and bottom scale features. Additionally, the model’s ability to focus on different parts of the image is improved by redesigning the EC2f structure and dynamically adjusting the convolutional kernel size to better capture features at various scales. Finally, the MHC2f mechanism is designed to enhance the model’s ability to perceive complex backgrounds and targets at different scales by utilizing its self-attention mechanism for parallel processing. Experiments demonstrate that the model’s mAP value increases by 4.1% compared to YOLOv8n, confirming its superiority in plant disease detection.