Most deep-learning-based object detection algorithms exhibit low speeds and accuracy in gear surface defect detection due to their high computational costs and complex structures. To solve this problem, a lightweight model for gear surface defect detection, namely STMS-YOLOv5, is proposed in this paper. Firstly, the ShuffleNetv2 module is employed as the backbone to reduce the giga floating-point operations per second and the number of parameters. Secondly, transposed convolution upsampling is used to enhance the learning capability of the network. Thirdly, the max efficient channel attention mechanism is embedded in the neck to compensate for the accuracy loss caused by the lightweight backbone. Finally, the SIOU_Loss is adopted as the bounding box regression loss function in the prediction part to speed up the model convergence. Experiments show that STMS-YOLOv5 achieves frames per second of 130.4 and 133.5 on the gear and NEU-DET steel surface defect datasets, respectively. The number of parameters and GFLOPs are reduced by 44.4% and 50.31%, respectively, while the mAP@0.5 reaches 98.6% and 73.5%, respectively. Extensive ablation and comparative experiments validate the effectiveness and generalization capability of the model in industrial defect detection.

Urban traffic flow forecasting remains challenging under heterogeneous urban functions, dynamic environmental conditions, and temporary gaps in recent regional observations. Existing spatio-temporal models primarily rely on historical traffic flows and often lack explicit representations of regional structure and urban semantics. We propose VLM-enhanced Context-aware Prior-guided Prediction with Decoupled Semantic Refinement (VLM-CPPD), a multimodal framework that integrates traffic dynamics with urban morphology, environmental context, and vision-language semantic priors derived from remote-sensing imagery. We further construct Porto-MMTF, a grid-aligned multimodal dataset that integrates inflow/outflow sequences, weather conditions, temporal context, OpenStreetMap morphology, remote-sensing imagery, and structured semantics extracted offline by MM-RSTraj. The core framework combines morphology-conditioned environmental adaptation, peak-aware forecast refinement, and a lightweight semantic-refinement interface. MM-RSTraj is used exclusively for offline semantic construction and is fully decoupled from online forecasting. Beyond the core framework, we introduce a masked-history extension for regions whose recent traffic observations are temporarily unavailable. In this setting, target-region structural and semantic information is combined with current traffic dynamics from observed grids to support prediction. On Porto-MMTF, VLM-CPPD achieves an RMSE of 3.7485, an MAE of 1.5435, and a Top10 MAE of 6.3692. On TaxiNYC, after city-specific data alignment and retraining, VLM-CPPD reduces RMSE, MAE, and Top10 MAE by 6.30%, 4.99%, and 6.08%, respectively, relative to STDN.