Construction of a Small Model Based on Large Model Knowledge Distillation in Anomaly Behaviour Recognition for Intelligent Connected Vehicles

Myriad: a large multimodal model applying vision experts for industrial anomaly detection

A novel large multimodal model applying vision experts for industrial anomaly detection (abbreviated as Myriad), which treats conventional IAD models as VEs and converts their anomaly maps into lightweight prompts that steer a frozen Q-Former toward suspicious regions, while a compact low-rank adapter shapes features for IAD.

An Explainable LLM Agent Layer for Open-World Anomaly Detection in Oil Wells

Open-World Learning (OWL) pipelines for oil well anomaly detection have recently been shown to combine autoencoder-based detection, multiclass classification, and Mahalanobis-based novelty detection on the public 3W dataset. These pipelines answer \textit{what happened}, but they do not explain \textit{why the model believes it} or \textit{what the operator should do next}, and they do not put a human-readable name on the novelty clusters they discover. This paper evaluates a Large Language Model (LLM) agent layer placed downstream of the OWL pipeline, designed as a \textbf{companion} to the published upstream methods rather than a replacement. Using the Qwen3.5-397B-A17B Mixture-of-Experts model served via NVIDIA NIM, the agent receives structured sensor metrics and upstream classification or novelty assertions, and returns natural-language justifications, confidence-ranked critiques, and consolidated names for detected novelties. Across three studies spanning 989 real well-file segments from the 3W dataset, the agent achieved $35.1\%$ top-1 / $63.9\%$ top-3 (95\% CI [56.9, 70.4]) classification on all nine classes, $71.7\%$ top-2 validation [64.8, 77.6] with precision $0.91$ [0.84, 0.95] across 7 probed classes, and $89.7\%$ novelty detection [87.0, 91.9] with stable cluster naming on 5 of 7 hidden classes. The agent is not a standalone classifier. Its role is to: (1) confirm upstream decisions when sensor evidence supports them, (2) justify decisions in sensor-grounded language operators can audit, (3) flag disagreement when upstream labels are implausible, and (4) name novelties so that clustered unlabeled events arrive at the engineer with a consolidated human-readable label. The goal is to close the explainability gap that currently blocks deployment of OWL pipelines in operational settings.

A Self-Contained Traffic Anomaly Detection System based on Distillation Learning for Resource Constrained Edge Deployment

Road traffic anomalies pose a critical threat to public safety, with fatality risk increasing by approximately 2.6% per minute of delayed medical response. While deep learning-based detection systems have shown promise, prevailing approaches either depend on computationally intensive architectures or cloud-based inference pipelines, rendering them unsuitable for resource-constrained deployments. Edge-capable alternatives, where they exist, typically demand relatively powerful dedicated hardware. This paper presents a fully self-contained traffic anomaly detection system deployed on a Raspberry Pi Zero 2W (based on a microprocessor operating at 1GHz and no specialized processing unit), without reliance on any external compute resources. The system is built on a lightweight spatiotemporal architecture trained via knowledge distillation, compressing a transformer-based teacher into a compact convolution-based student. The deployed model achieves a Top-1 accuracy of 76.23%, operating at 16.74FPS with an inference latency of 59.75ms on the target hardware.

Multi domain neural fusion for adaptive anomaly detection in connected and automated vehicles

Connected and Automated Vehicles (CAVs) rely on high-dimensional, multimodal sensor data and Vehicle-to-Everything (V2X) communication to support autonomous driving functions. This strong dependence exposes CAV systems to anomalies arising from sensor faults, environmental disturbances, and coordinated cyber–physical attacks. To address these challenges, this paper proposes a real-time hybrid anomaly detection framework that integrates multiple domain-specific models. The framework combines an LSTM-based time-domain model for sequential behavior analysis, an FFT-based frequency-domain model for spectral anomaly detection, and a context-aware multilayer perceptron (MLP) for incorporating environmental factors. To overcome data scarcity and improve robustness, synthetic anomalies are generated and integrated with benchmark datasets, including NGSIM, CAN-ID, and nuScenes. A learned fusion layer aggregates the outputs of the individual models into a unified anomaly score, while an online learning module dynamically adapts detection thresholds using sliding-window percentile recalibration. In addition, a Deep Q-Network (DQN)–based reinforcement learning agent supports adaptive decision-making under changing conditions. Model interpretability is enhanced through SHAP-based explanations and Human-in-the-Loop feedback, enabling continuous refinement of the detection process. Experimental results show that the proposed system achieves an accuracy of 98.10% before adaptation and improves to 98.91% after online learning and feedback. These results demonstrate that the proposed framework offers strong robustness, adaptability, and real-time suitability for safety-critical CAV environments.

AnomalyMHKD: A multi-mode hybrid knowledge distillation approach for self-supervised detection of anomalous objects

OPD-IAD: From Language Judgment to Industrial Anomaly Detection via On-Policy Self-Distillation

Large vision-language models (LVLMs) have recently shown strong potential for industrial anomaly detection (IAD) by providing image-level anomaly judgments and interpretable defect reasoning. However, current LVLM-based IAD methods still struggle to produce precise pixel-level anomaly maps from generated language judgments. We aim to achieve precise pixel-level localization while using language as guidance rather than letting it dominate the visual response. Specifically, we propose \textbf{OPD-IAD}, an evidence-privileged dense on-policy self-distillation framework for LVLM-based IAD. OPD-IAD distills privileged defect evidence onto the model's own on-policy judgment trajectory, enabling the final generated judgment to be learned under dense supervision rather than treated only as a textual answer. The resulting judgment serves as a semantic condition for dense anomaly perception. To turn this condition into dense visual evidence, we introduce \textbf{Language-guided Visual Anchoring}, which uses a judgment reforward to re-encode the image and question under the final-judgment condition into semantic anchors and contrasts them with dense visual features through a contrastive heatmap head to generate anomaly maps. The language judgment therefore provides compact semantic guidance, while dense visual features remain the basis for pixel-level scoring, allowing language to guide anomaly localization without letting language quality directly dictate the pixel-level response. Extensive experiments show that OPD-IAD achieves the best overall performance among LVLM-based IAD methods, leading on most image-level, pixel-level, and QA metrics.