Forest fires rank among the costliest and deadliest natural disasters globally. Identifying the smoke generated by forest fires is pivotal in facilitating the prompt suppression of developing fires. Nevertheless, succeeding techniques for detecting forest fire smoke encounter persistent issues, including a slow identification rate, suboptimal accuracy in detection, and challenges in distinguishing smoke originating from small sources. This study presents an enhanced YOLOv8 model customized to the context of unmanned aerial vehicle (UAV) images to address the challenges above and attain heightened precision in detection accuracy. Firstly, the research incorporates Wise-IoU (WIoU) v3 as a regression loss for bounding boxes, supplemented by a reasonable gradient allocation strategy that prioritizes samples of common quality. This strategic approach enhances the model’s capacity for precise localization. Secondly, the conventional convolutional process within the intermediate neck layer is substituted with the Ghost Shuffle Convolution mechanism. This strategic substitution reduces model parameters and expedites the convergence rate. Thirdly, recognizing the challenge of inadequately capturing salient features of forest fire smoke within intricate wooded settings, this study introduces the BiFormer attention mechanism. This mechanism strategically directs the model’s attention towards the feature intricacies of forest fire smoke, simultaneously suppressing the influence of irrelevant, non-target background information. The obtained experimental findings highlight the enhanced YOLOv8 model’s effectiveness in smoke detection, proving an average precision (AP) of 79.4%, signifying a notable 3.3% enhancement over the baseline. The model’s performance extends to average precision small (APS) and average precision large (APL), registering robust values of 71.3% and 92.6%, respectively.

Authorities and policymakers in Korea have recently prioritized improving fire prevention and emergency response. Governments seek to enhance community safety for residents by constructing automated fire detection and identification systems. This study examined the efficacy of YOLOv6, a system for object identification running on an NVIDIA GPU platform, to identify fire-related items. Using metrics such as object identification speed, accuracy research, and time-sensitive real-world applications, we analyzed the influence of YOLOv6 on fire detection and identification efforts in Korea. We conducted trials using a fire dataset comprising 4000 photos collected through Google, YouTube, and other resources to evaluate the viability of YOLOv6 in fire recognition and detection tasks. According to the findings, YOLOv6’s object identification performance was 0.98, with a typical recall of 0.96 and a precision of 0.83. The system achieved an MAE of 0.302%. These findings suggest that YOLOv6 is an effective technique for detecting and identifying fire-related items in photos in Korea. Multi-class object recognition using random forests, k-nearest neighbors, support vector, logistic regression, naive Bayes, and XGBoost was performed on the SFSC data to evaluate the system’s capacity to identify fire-related objects. The results demonstrate that for fire-related objects, XGBoost achieved the highest object identification accuracy, with values of 0.717 and 0.767. This was followed by random forest, with values of 0.468 and 0.510. Finally, we tested YOLOv6 in a simulated fire evacuation scenario to gauge its practicality in emergencies. The results show that YOLOv6 can accurately identify fire-related items in real time within a response time of 0.66 s. Therefore, YOLOv6 is a viable option for fire detection and recognition in Korea. The XGBoost classifier provides the highest accuracy when attempting to identify objects, achieving remarkable results. Furthermore, the system accurately identifies fire-related objects while they are being detected in real-time. This makes YOLOv6 an effective tool to use in fire detection and identification initiatives.

Agricultural fires significantly threaten global food systems, ecosystems, and rural economies, necessitating timely detection to prevent widespread damage. This study presents a lightweight and enhanced version of the YOLOv5s model, optimized for early-stage agricultural fire detection. The core innovation involves deepening the C3 block and integrating DarknetBottleneck modules to extract finer visual features from subtle fire indicators such as light smoke and small flames. Experimental evaluations were conducted on a custom dataset of 3200 annotated agricultural fire images. The proposed model achieved a precision of 88.9%, a recall of 85.7%, and a mean Average Precision (mAP) of 87.3%, outperforming baseline YOLOv5s and several state-of-the-art (SOTA) detectors such as YOLOv7-tiny and YOLOv8n. The model maintains a compact size (7.5 M parameters) and real-time capability (74 FPS), making it suitable for resource-constrained deployment. Our findings demonstrate that focused architectural refinement can significantly improve early fire detection accuracy, enabling more effective response strategies and reducing agricultural losses.

Diagnosing faults across different smart homes is hard: sensor names, layouts, and daily routines differ from home to home, so a model trained in one home rarely works in another. We study an ontology-guided framework for cross-home fault diagnosis, but our main contribution is a benchmark-validity audit—a systematic check of whether the datasets used to evaluate such systems actually measure fault detection. Using the public Center for Advanced Studies in Adaptive Systems (CASAS) smart-home datasets (homes hh101–hh110) and real household power data (HomeC, UMass Smart*), we show that much of the high cross-home accuracy reported on these benchmarks is an artifact of features that re-encode the labelling rules rather than evidence of transfer: when those features are removed, the macro-averaged F1 score (macro-F1) collapses toward the level obtained with randomly permuted labels. We therefore treat these datasets as semantic-transfer and benchmark-validity studies, not fault-detection results. The framework’s distinguishing component is a counterfactual calibration layer that returns a probability for its recommended intervention; on a controlled structural causal model with known interventions, it achieves a Brier skill score of $0.369$ for intervention-success probabilities. Separately, on the simulation-derived LBNL Fan Coil Unit benchmark, a conventional gradient-boosted multiclass fault classifier achieves accuracy comparable to a random forest but about six times lower expected calibration error ($0.026$ vs. $0.159$) under a scenario-matched split. This calibration advantage does not generalize to held-out simulation scenarios, where the calibration error rises to $0.372$; we report this negative result as a limitation. We are explicit about scope: the ontology reasoner and the real-stream causal graph are only partially implemented, and the counterfactual recommendations are validated only under controlled or simulated conditions, not in deployed homes. The results are intended for researchers who build or benchmark sensor-based fault-diagnosis models, for dataset curators, and for practitioners who need calibrated rather than merely accurate outputs. All code, the proxy-label rules, and the leakage audit are released.

Which runtime governance mechanisms reduce physical risk when a learned controller drives a building’s climate, and under what conditions? We develop a closed-loop software-in-the-loop testbed in which a setpoint model trained on real occupancy data drives a physics-based thermal zone through a declarative governance plane, with outcomes scored by an independent safety oracle, and we run the same governance logic on a real MQTT stack with an in-process policy decision point and a hash-chained audit log. Under distribution shift, admission control reduces unsafe physical exposure by $19.4\%$, from 1185.3 to 954.8 °C·min, whereas adding checkpoint rollback reduces it by only a further $0.2\%$ in the reference run ($0.1$–$0.4\%$ across sensor noise seeds): the governance decision takes 0.44 ms while physical recovery takes a median of 61 min. Prevention therefore outperforms recovery in the studied thermal system, and the remaining avoidable exposure is driven by the policy’s estimate of occupancy context. A deterministic single-rule thermostat incurs $50\%$ more exposure under shift while the learned controller uses a $38\%$ higher heating demand proxy: a safety–demand trade-off, not evidence that learned control is necessary. A plant sweep yields an operating envelope criterion for inertial plants: rollback contributes materially to safety only when the plant is restored before the next command arrives and sampled before it can leave the safe set; on slower plants it removes at most $14.7\%$, with a transition band in between. No physical hardware was operated.

We ask whether an artificial intelligence (AI)-enabled process monitor’s alarms remain informative beyond its commissioned operating range. Conformal alarm rules rely on exchangeability, which deployment shifts can break. Conditioning calibration on the operating point restores validity, not necessarily informativeness: the certificate constrains only the fault-free calibration law, so a diagnostic relying only on that law cannot identify the loss without an additional assumption. Across ten one-class detectors in a seed-controlled Tennessee Eastman campaign with counterfactual twins, validity holds at every evaluated level while informativeness decays. Resolved at $\times 0.001$, the loss is a ramp, not a step, crossing a declared $0.70$ floor at $\times 0.825$. There the plant is bistable: fault-free episodes settle into a steady mode or an unsteady one with the purge valve on its stop, the detectors cannot rank the fault inside the unsteady mode, and informativeness above chance scales with the square of the steady share. Disabling that protective override removes the unsteady mode and lifts far-edge receiver operating characteristic (ROC) area from $0.47$ to $0.76$. Supervised classifiers recover the ranking only when trained on labelled faults in that settled state. Fault-free commissioning measurements, mapped by a curve fitted on labelled faults, predict the loss to a mean absolute ROC-area error of $0.041$.