Selected work

Representative Papers

Foreign-Object Detection in High-Voltage Transmission Line Based on Improved YOLOv8m

Nov 28, 2023Applied Sciences

To address low detection accuracy in identifying foreign objects (e.g., balloons, kites, bird nests) on high-voltage transmission lines—caused by severe occlusion, large scale variations, and complex backgrounds—this paper proposes a lightweight and efficient detection method based on an improved YOLOv8m architecture. The method introduces three key innovations: (1) a Global Attention Mechanism (GAM) to enhance perception of occluded targets; (2) SPPCSPC replacing SPPF to improve multi-scale feature fusion efficiency; and (3) a Focal-EIoU loss function to mitigate positive–negative sample imbalance. Evaluated on a real-world dataset collected from Yunnan Power Grid, the proposed model achieves +2.7% mAP₀.₅, +4.0% mAP₀.₅:₀.₉₅, and +6.0% recall over the baseline. These improvements significantly enhance robustness and practicality for detecting small and occluded foreign objects under challenging field conditions.

17 citations1 influentialRead paper

Improved YOLOv5s model for key components detection of power transmission lines.

Feb 20, 2023Mathematical biosciences and engineering : MBE

To address the low detection accuracy of small critical components (e.g., insulators, hardware) in complex backgrounds during intelligent inspection of high-voltage transmission lines, this paper proposes a lightweight and efficient YOLOv5s-based detection method. Specifically, we: (i) design an IoU-optimized k-means clustering distance metric for anchor box generation; (ii) incorporate the Convolutional Block Attention Module (CBAM) to enhance multi-scale feature discrimination; and (iii) adopt Focal Loss to mitigate class imbalance among component categories. Experimental results on a custom-built transmission line dataset show that the proposed model achieves 98.1% mAP, 97.5% precision, and 94.4% recall, with an inference speed of 84.8 FPS—significantly outperforming the baseline YOLOv5s. The method thus delivers both high accuracy and real-time performance, effectively supporting practical deployment in intelligent transmission line inspection systems.

9 citations2 influentialRead paper

A Multi-View and Confusion-Guided Ensemble Framework for Robust Synthetic Image Attribution

Sep 10, 2026

本文针对合成图像归因难题，提出一个多视图与混淆引导集成框架，结合多种架构捕捉不同线索，并通过数据增强和专用分类器提高鲁棒性。

0 citationsRead paper

FMOPF: Latent Flow Matching with Constraint-Aware Interaction Priors for AC Optimal Power Flow

Jul 24, 2026

This work addresses the challenge of efficiently modeling the conditional distribution of near-optimal feasible solutions to AC optimal power flow (AC OPF) under high renewable energy penetration, where conventional methods struggle. The authors propose a decoupled framework that first maps the high-dimensional OPF solution manifold into a latent space via neural compression, then explicitly models the conditional mapping from load inputs to the solution manifold using latent-space flow matching combined with a constraint-aware interactive prior network. This approach uniquely decouples the compression and generation tasks. Validated on four IEEE test systems, the method consistently generates fully feasible solutions with the lowest tail risk among compared techniques, scales effectively to systems with hundreds of buses, and provides high-quality initializations for Newton-Raphson solvers.

0 citationsRead paper

DenoiseFlow: Uncertainty-Aware Denoising for Reliable LLM Agentic Workflows

Feb 28, 2026

This work addresses the degradation in reliability of large language models within long-sequence agent workflows caused by accumulating semantic ambiguities. To mitigate this, the authors propose a three-stage closed-loop framework that models multi-step reasoning as a noisy Markov decision process, enabling progressive denoising through perception, regulation, and correction. Key innovations include an uncertainty-aware adaptive computation allocation mechanism, an unsupervised online self-calibration method, and a synergistic integration of semantic uncertainty estimation, adaptive path exploration, impact-analysis-driven error correction, and verifier feedback alignment. Evaluated across six benchmarks, the approach achieves an average accuracy of 83.3%, outperforming the strongest baseline by 1.3%, while reducing computational overhead by 40–56% through dynamic branching.

0 citationsRead paper

Recent publications

Latest Papers

A Multi-View and Confusion-Guided Ensemble Framework for Robust Synthetic Image Attribution

Sep 10, 2026

本文针对合成图像归因难题，提出一个多视图与混淆引导集成框架，结合多种架构捕捉不同线索，并通过数据增强和专用分类器提高鲁棒性。

0 citationsRead paper

FMOPF: Latent Flow Matching with Constraint-Aware Interaction Priors for AC Optimal Power Flow

Jul 24, 2026

This work addresses the challenge of efficiently modeling the conditional distribution of near-optimal feasible solutions to AC optimal power flow (AC OPF) under high renewable energy penetration, where conventional methods struggle. The authors propose a decoupled framework that first maps the high-dimensional OPF solution manifold into a latent space via neural compression, then explicitly models the conditional mapping from load inputs to the solution manifold using latent-space flow matching combined with a constraint-aware interactive prior network. This approach uniquely decouples the compression and generation tasks. Validated on four IEEE test systems, the method consistently generates fully feasible solutions with the lowest tail risk among compared techniques, scales effectively to systems with hundreds of buses, and provides high-quality initializations for Newton-Raphson solvers.

0 citationsRead paper

DenoiseFlow: Uncertainty-Aware Denoising for Reliable LLM Agentic Workflows

Feb 28, 2026

This work addresses the degradation in reliability of large language models within long-sequence agent workflows caused by accumulating semantic ambiguities. To mitigate this, the authors propose a three-stage closed-loop framework that models multi-step reasoning as a noisy Markov decision process, enabling progressive denoising through perception, regulation, and correction. Key innovations include an uncertainty-aware adaptive computation allocation mechanism, an unsupervised online self-calibration method, and a synergistic integration of semantic uncertainty estimation, adaptive path exploration, impact-analysis-driven error correction, and verifier feedback alignment. Evaluated across six benchmarks, the approach achieves an average accuracy of 83.3%, outperforming the strongest baseline by 1.3%, while reducing computational overhead by 40–56% through dynamic branching.

0 citationsRead paper

Buffer Matters: Unleashing the Power of Off-Policy Reinforcement Learning in Large Language Model Reasoning

Feb 24, 2026

This work addresses the inefficiencies of online policy reinforcement learning in traditional large language model (LLM) post-training, which often leads to wasted experience and reward homogenization, hindering effective learning from challenging samples. To overcome these limitations, the authors propose Batch-Adaptive Policy Optimization (BAPO), a novel framework that introduces off-policy reinforcement learning into LLM post-training for the first time. BAPO employs a replay buffer mechanism to dynamically re-evaluate and reuse high-value historical trajectories, thereby enhancing data efficiency while providing a theoretical lower bound on policy performance improvement. Experimental results demonstrate that BAPO outperforms GRPO by an average of 12.5% across mathematical reasoning, planning, and visual reasoning tasks, and successfully solves 40.7% of problems that were previously unsolvable by the base model.

0 citationsRead paper

The Energy of Falsehood: Detecting Hallucinations via Diffusion Model Likelihoods

Feb 11, 2026

This work proposes a novel thermodynamic perspective for detecting hallucinations in large language models, which often generate highly confident yet factually incorrect statements that evade conventional uncertainty metrics. Treating factual claims as stable attractors on the generative manifold and hallucinations as unstable states, the method leverages a discrete text diffusion model to perturb and reconstruct input claims. A natural language inference (NLI) discriminator is introduced to compute semantic energy, quantifying deep semantic inconsistencies between original and reconstructed statements. This approach uniquely integrates thermodynamic stability into hallucination detection by jointly calibrating generative stability and discriminative confidence. Evaluated without supervision, it achieves an AUROC of 0.725 on the FEVER dataset, outperforming baselines by 1.5%, and demonstrates over 4% zero-shot improvement on the multi-hop HOVER benchmark, confirming robustness to distributional shifts.

0 citationsRead paper