Rose tea is a type of flower tea in China’s reprocessed tea category, which is divided into seven grades, including super flower, primary flower, flower bud, flower heart, yellow flower, scattered flower, and waste flower. Grading rose tea into distinct quality levels is a practice that is essential to boosting their competitive advantage. Manual grading is inefficient. We provide a lightweight model to advance rose tea grading automation. Firstly, four kinds of attention mechanisms were introduced into the backbone and compared. According to the experimental results, the Convolutional Block Attention Module (CBAM) was chosen in the end due to its ultimate capacity to enhance the overall detection performance of the model. Second, the lightweight module C2fGhost was utilized to change the original C2f module in the neck to lighten the network while maintaining detection performance. Finally, we used the SIoU loss in place of the CIoU loss to improve the boundary regression performance of the model. The results showed that the mAP, precision (P), recall (R), FPS, GFLOPs, and Params values of the proposed model were 86.16%, 89.77%, 83.01%, 166.58, 7.978, and 2.746 M, respectively. Compared with the original model, the mAP, P, and R values increased by 0.67%, 0.73%, and 0.64%, the GFLOPs and Params decreased by 0.88 and 0.411 M, respectively, and the speed was comparable. The model proposed in this study also performed better than other advanced detection models. It provides theoretical research and technical support for the intelligent grading of roses.

This study proposes a novel method that combines multispectral data fusion strategies with chemometric analysis for the origin traceability of Rizhao green tea. The research found significant differences in the sensory scores and key physicochemical components (catechins, caffeine, and amino acid content) between Rizhao green tea and tea from southern China. By integrating data from near-infrared and hyperspectral technologies, the prediction accuracy of multivariate models (including Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest (RF), and Convolutional Neural Networks (CNN)) was improved. The performance of the fused dataset outperformed single-spectral datasets. The study found significant spectral differences in tea samples from different regions, leading to robust differentiation. Both SVM and RF discriminant models based on near-infrared spectral data achieved 100 % accuracy. This method provides a reliable and efficient tool for green tea traceability, with potential applications in quality control and authenticity verification within the tea industry.

Focusing on the characteristic tea resource Zijuan tea, this study addresses the difficulty of grading on production lines and the complexity of quality evaluation. On the basis of the fusion of near-infrared (NIR) spectroscopy and visual features, a novel method is proposed for classifying different tenderness levels and quantitatively assessing key anthocyanin components in Zijuan tea fresh leaves. First, NIR spectra and visual feature data were collected, and anthocyanin components were quantitatively analyzed using UHPLC-Q-Exactive/MS. Then, four preprocessing techniques and three wavelength selection methods were applied to both individual and fused datasets. Tenderness classification models were developed using Particle Swarm Optimization–Support Vector Machine (PSO-SVM), Random Forest (RF), and Convolutional Neural Networks (CNNs). Additionally, prediction models for key anthocyanin content were established using linear Partial Least Squares Regression (PLSR), nonlinear Support Vector Regression (SVR) and RF. The results revealed significant differences in NIR spectral characteristics across different tenderness levels. Model combinations such as TEX + Medfilt + RF and NIR + Medfilt + CNN achieved 100% accuracy in both training and testing sets, demonstrating robust classification performance. The optimal models for predicting key anthocyanin contents also exhibited excellent predictive accuracy, enabling the rapid and nondestructive detection of six major anthocyanin components. This study provides a reliable and efficient method for intelligent tenderness classification and the rapid, nondestructive detection of key anthocyanin compounds in Zijuan tea, holding promising potential for quality control and raw material grading in the specialty tea industry.

Introduction: Rapid assessment of soil pH and nutrient status in tea plantations is essential for precision fertilisation and ecological management, particularly in high-latitude tea-growing regions where related applications remain insufficiently studied. Methods: This study developed a rapid, non-destructive framework integrating hyperspectral imaging and machine learning for the detection, retrieval, and spatial visualisation of soil pH, soil organic matter (SOM), alkali-hydrolysable nitrogen (AN), available phosphorus (AP), and available potassium (AK) in the Taishan tea-producing region. A total of 150 soil samples were collected from three profile depths (0–20, 20–40, and 40–60 cm). Hyperspectral images were acquired over 394–1007 nm, and the 481–908 nm range was retained for modelling. Principal component analysis was used to characterise vertical differentiation, while four spectral pre-processing methods, three feature-band selection algorithms—competitive adaptive reweighted sampling (CARS), bootstrapping soft shrinkage (BOSS), and successive projections algorithm (SPA)—and three regression models—partial least squares regression (PLSR), random forest (RF), and support vector regression (SVR)—were systematically compared. Results: The soils were generally acidic, and SOM, AN, AP, and AK exhibited clear surface enrichment and decreasing trends with increasing depth. Among the feature-selection methods, BOSS showed the best overall performance in reducing spectral redundancy and improving prediction accuracy. The optimal SVR models, combined with parameter-specific pre-processing and BOSS-selected bands, achieved strong predictive performance across all indicators, with prediction-set correlation coefficients (Rp) of 0.94–0.99 and relative percent deviation (RPD) values of 2.963–10.425. Furthermore, pixel-wise reconstruction using threshold masking enabled intuitive two-dimensional visualisation of the spatial distributions of the target soil properties. Discussion: These results demonstrate that hyperspectral imaging coupled with machine learning provides an effective approach for rapid soil nutrient assessment, spatial visualisation, and digital management in high-latitude tea plantations.