Reliable operating-state recognition of gas pressure regulators is essential for pressure stability, operational safety, and supply continuity in urban natural gas distribution networks. However, SCADA pressure–flow signals from regulating stations are often affected by non-stationary noise, impulsive disturbances, limited fault-state samples, and short-window temporal fluctuations, which reduce the reliability of data-driven recognition. To address these issues, this study proposes K²-TLNet, a noise-state-guided fault-state recognition framework for gas pressure regulation processes. The framework integrates adaptive Kalman filtering, training-only KMeans-SMOTE, parallel temporal convolutional network–long short-term memory feature extraction, and a Noise-Aware Gated Fusion mechanism. Adaptive Kalman filtering is used to generate denoised pressure–flow sequences and extract innovation-residual-based noise descriptors. These descriptors guide the fusion module to adaptively balance local transient features from the temporal convolutional network and contextual temporal features from long short-term memory. A field-SCADA-background-based semi-synthetic dataset was constructed using real operating records and mechanism-informed fault-state emulation rules. Experimental results demonstrate that K²-TLNet achieves 98.50% accuracy and 98.20% Macro-F1, while maintaining strong robustness under Gaussian and impulsive noise disturbances.

Weed management is an important part of agricultural production and a key factor affecting crop yields. However, weeds in the field have characteristics such as mixed categories, high similarity in morphology with crops, and small detection targets, making detection quite difficult. To address these issues, this paper proposes an improved lightweight model, MBP-RTDETR, based on the RTDETR model. This is a real-time detection algorithm that reduces computational load while effectively enhancing the detection accuracy of small-sized weeds in complex farmland backgrounds. The key improvements include: (1) MSNet, a lightweight backbone network, integrates the MSInit module into the CSPNet architecture, effectively incorporating contextual information to provide enhanced multi-scale feature extraction at low computational cost. (2) The AIFI-SSA module simplifies multiplication into efficient binary masking and accumulation operations by leveraging the binary characteristics of peaks, reducing the number of model parameters while maintaining detection accuracy. (3) The PST-DET architecture replaces the conventional convolutional layers in the Feature Pyramid Network (FPN) with the PST module, thereby reducing computational complexity while preserving critical spatial information. Experimental results demonstrate that the proposed model achieves mAP@50 and mAP@50:95 scores of 93.7% and 75.2%, respectively, outperforming the baseline model across both evaluation metrics. Importantly, this performance gain is accompanied by significant efficiency improvements: the model reduces parameter count, computational complexity (GFLOPs), and model weight by 41.2%, 33.3%, and 41.0%, respectively. These enhancements collectively enable robust and computationally efficient detection of field weeds under complex real-world conditions.