The prediction models for tunnel fires require exceptionally high accuracy and responsiveness due to the complex enclosed environment and associated high-risk characteristics. However, existing temporal prediction models demonstrate limitations in processing multidimensional data and managing long-term dependencies, making them inadequate for effectively addressing the prediction requirements of tunnel fires. This study introduces the bidirectional down-sampling informer (BiDS-Informer) model, which is built upon innovative components, including the BiDS module, dynamic attention mechanisms, and encoder dimensionality reduction optimization. (1) Within the encoder, the BiDS-Informer utilizes the bidirectional down-sampling module for feature extraction, thereby obtaining a more comprehensive representation of features; (2) the traditional self-attention mechanism is restructured into dynamic attention mechanisms, allowing the model to adaptively identify and process the characteristics of multidimensional time series data; (3) finally, dimensionality reduction is applied to the encoder’s input structure to eliminate temporal encoding information, resulting in enhanced computational efficiency. Experimental results indicate that the proposed model outperforms existing time series prediction models regarding predictive accuracy across multiple real-world datasets, particularly on the tunnel fire dataset. Compared to the informer model, the proposed model shows improvements of 18.56% and 27.49% in average MSE and MAE, respectively, on the tunnel fire dataset. This model provides substantial technical support for enhancing the reliability of tunnel fire early warning systems and holds promising potential for widespread application.

Currently, train component fault detection is predominantly carried out through manual inspection, a process that is inefficient, prone to high omission rates, and carries safety risks. This study proposes an innovative fault detection model for train components based on YOLOv8, aiming to overcome the inefficiencies and high omission rates associated with traditional manual methods. By optimizing the YOLOv8 network architecture and integrating the ADown module, C2f-Rep, and DHD, the model significantly improves computational efficiency and detection accuracy. Experimental results demonstrate that the optimized Train-YOLO model achieves a peak accuracy of 92.9% in train component fault detection. Additionally, it features a smaller model size and reduced computational demands, making it ideal for rapid on-site deployment. A comparison with other leading detection models further highlights the superiority of Train-YOLO in both accuracy and lightweight design.

Accurate weed detection in crop fields remains a challenging task due to the diversity of weed species and their visual similarity to crops, especially under natural field conditions where lighting and occlusion vary. Traditional methods typically attempt to directly identify various weed species, which demand large-scale, finely annotated datasets and often suffer from low generalization. To address these challenges, this study proposes a novel dual-model framework that simplifies the task by dividing it into two tractable stages. First, a crop segmentation network is used to identify and remove cabbage (Brassica oleracea L. ssp. pekinensis) regions from field images. Since crop categories are visually consistent and singular, this stage achieves high precision with relatively low complexity. The remaining non-crop areas, which contain only weeds and background, are then subdivided into grid cells. Each cell is classified by a second lightweight classification network as either background, broadleaf weeds, or grass weeds. The classification model achieved F₁ scores of 95.1%, 91.1%, and 92.2% for background, broadleaf weeds, and grass weeds, respectively. This two-stage approach transforms a complex multi-class detection task into simpler, more manageable subtasks, improving detection accuracy while reducing annotation burden and enhancing robustness under the tested field conditions.

With the continuous advancement of underground tunnel systems, the academic community has shown increasing concern about the hazards caused by water leakage.As one of the current research directions, instance segmentation algorithms still face several challenges in detecting water leakage defects in underground tunnels, such as insufficient accuracy in detecting small-sized targets and the large size of highprecision models, which is unfavorable for practical deployment. In view of this, this paper aims to design an instance segmentation model that maintains high accuracy while incorporating lightweight characteristics.This paper employs the Mosaic data augmentation technique for image enhancement, rendering image features more prominent and thus more favorable for machine learning.At the model level, we introduce a Local Importance-based Attention (LIA) mechanism, which aims to reduce computational complexity while preserving model performance, thereby enabling efficient information interaction. We further integrate the CAS_ViT and C3K2 modules into a unified C3K2_CAS module to replace the original C3K2 module-this design effectively captures fine-grained features while maintaining computational tractability. To further refine local features and enhance spatial information, we incorporate a C3K2_CAFormer module, which enhances local details such as edges while suppressing background noise.The optimized model achieves a Box_mAP50 of 80.3% and a Mask_mAP50 of 85.8% with only 3.2 million parameters. Compared to various mainstream models, it demonstrates the best overall performance, providing a low-cost and highly efficient solution for detecting water leakage defects in underground tunnels.

Traditional methods for wine bottle packaging leakage detection often suffer from low efficiency, high false-positive rates, or an inability to detect micro-leakages. This paper proposes a near-infrared laser leakage detection system based on tunable diode laser absorption spectroscopy at 1392 nm, combined with a LightGBM machine learning model. The system detects gaseous ethanol vapor escaping from leaking bottles, addressing the spectral interference caused by ambient water vapor. A total of 1410 samples were collected, and each raw 2000-point spectral contour was compressed into a 200-dimensional feature vector through baseline correction, Z-score normalization, and uniform down-sampling. A two-stage hyperparameter optimization strategy yielded the optimal LightGBM configuration with a 5-fold cross-validation. For the binary classification task, the model achieved an AUC of 0.9949 and an inference speed of 0.0058 ms per sample on a CPU, outperforming Random Forest, PLS, and four deep learning models. For the regression task, the model achieved an R² of 0.5854 ± 0.0919. An anti-interference experiment on 422 samples under varying flow rates, temperatures, and commercial wine types confirmed the model’s robustness, achieving an overall accuracy of 0.94 and an alcohol recall of 0.99. To further validate the system under realistic conditions, a simulated micro-leakage test was conducted using a negative-pressure extraction method: 320 samples were collected from artificially damaged commercial wine bottles placed in a custom-built acrylic vacuum chamber that replicates the production line enclosure. The model achieved an accuracy of 0.95 with zero false negatives. The complete detection cycle takes no more than 5 s per bottle, enabling non-destructive, rapid, and online packaging integrity assessment. The results demonstrate that the proposed system provides a low-cost and reliable solution for wine bottle leakage detection suitable for industrial deployment.

Accurate forecasting of multi-channel parameters—temperature, carbon monoxide concentration, and wind speed—is essential for emergency ventilation and evacuation during tunnel fires. Fire temporal curves, however, couple slow convective evolution with abrupt, localized fluctuations driven by transient physical disturbances, and existing deep learning models struggle to separate these multi-scale dependencies, causing errors to accumulate over long horizons. We propose PatchTST-multilayer perceptron (MLP), a hybrid model for multi-channel tunnel fire forecasting. The architecture uses patch-based embedding to capture local temporal context at low computational cost. A context-aware decomposition mechanism then splits the embedded sequence into a long-term trend and a transient-disturbance component; the Transformer encoder models global long-range dependencies while the MLP encoder captures high-frequency fluctuations. We constructed a full-scale tunnel fire dataset of 111 cross-sectional channels under varied heat release rates and ventilation conditions using fire dynamics simulator, with a strict sequential partitioning strategy to prevent data leakage. PatchTST-MLP achieves competitive or superior performance against state-of-the-art time-series baselines across short- and long-term horizons. Ablation studies show that 2 Transformer blocks and 2 MLP layers give the best accuracy–complexity trade-off making it practical for real-time tunnel fire safety management.