Volatile organic compounds (VOCs) are closely associated with the maturity and variety of strawberries. However, the complexity of VOCs hinders their potential application in strawberry classification. This study developed a novel classification workflow using strawberry VOC profiles and machine learning (ML) models for precise fruit classification. A comprehensive VOC dataset was rapidly collected using gas chromatography-ion mobility spectrometry (GC-IMS) from five strawberry varieties at four maturity stages (n = 300) and visualized through principal component analysis (PCA). Five ML models were developed, including partial least squares discriminant analysis (PLS-DA), decision trees, support vector machines (SVM), Xgboost and neural networks (NN). The accuracy of all models ranged from 90.00% to 98.33%, with the NN model demonstrating the best performance. Specifically, it achieved 96.67% accuracy for single-maturity classification, 98.33% for single-variety classification, and 96.67% for dual maturity and variety classification, along with 98.09% precision, 97.92% recall, and 97.91% F1 score. Feature importance analysis indicated that the NN model exhibited the most balanced reliance on various VOCs, contributing to its optimal performance with the broad-spectrum VOC detection method, GC-IMS. Overall, these findings underscore the potential of NN modeling for accurate and efficient fruit classification based on integrated VOC profiles.

Traditional tea quality evaluation depends on human evaluators, limiting scalability, and consistency. To establish an artificial intelligence (AI)‐assisted framework for comprehensive evaluation of tea quality and detailed assessment of the tea consumption experience, this study aims to develop Long‐Tea‐CLIP (Contrastive Language‐Image Pre‐training), a multimodal tea grading system that combines computer vision and chemoinformatics. It integrates five sensory evaluation dimensions for green tea, using separate submodels for appearance (ResNet‐18), soup color (eXtreme Gradient Boosting (XGBoost)), aroma, infused leaf, and taste (multilayer perceptron (MLP)). A deep network derived from ResNet‐18 integrates dry tea images with seven subdimensions of sensory comments to achieve a refined appearance “grading.” We apply Tip‐CLIP supervised MLP on feature data extraction from infused leaf and chemical data of aroma and taste to enhance accuracy. Submodel outputs are weighted into a unified framework to produce an overall score. Long‐Tea‐CLIP trained on 7763 image‐text pairs from 38 Longjing tea varieties achieves 92% accuracy, indicating its potential to enhance tea quality control and market transparency.

The Traveling Salesman Problem (TSP) is a representative NP-hard combinatorial optimization problem whose search space grows rapidly with the number of cities. Classical ant colony optimization (ACO) is attractive because of its distributed construction process and positive-feedback search, but its performance deteriorates on large instances owing to probability dilution, pheromone concentration, search stagnation, and the cost of optimizing a complete tour with a single colony. This paper proposes Collaborative Attention Ant Colony Optimization (CAACO), which integrates sparse coordinate-based graph attention, random-offset collaborative subpath optimization, a memetic MAX–MIN Ant System, and adaptive double-bridge stagnation escape. The method is evaluated against FACO, BCACO, and CCACO on nine symmetric Euclidean TSP instances, with every algorithm executed independently 30 times per instance. CAACO obtains the lowest mean GAP on all nine instances, with an average instance-level mean GAP of 3.193%, compared with 12.587%, 22.177%, and 23.826% for the three baselines. Per-instance two-sided Mann–Whitney U tests with Holm correction show significant differences in all 27 comparisons (pHolm < 10⁻¹⁰), and the Vargha–Delaney effect size is A₁₂=1.000 throughout. These results demonstrate a consistently favorable solution-quality distribution for the proposed framework, while the runtime analysis shows a quality–efficiency trade-off relative to the efficiency-oriented baseline. A 30-run ablation study further shows that removing attention, dynamic slicing, and double-bridge escape increases the average mean GAP from 3.199% to 4.525%, 5.863%, and 3.813%, respectively; all 27 module-level comparisons remain significant after Holm correction.