In Internet of Things (IoT) scenarios, vision-language models (VLMs) are increasingly employed for visual perception and reasoning. However, their inherent tendency toward hallucinated and post-hoc rationalized reasoning often produces false positives (FPs), where the final answer is correct but supported by unreliable reasoning paths, typically manifested as self-inconsistent reasoning. Existing methods primarily focus on forward enhancement of reasoning ability, while neglecting reverse verification of reasoning consistency. To address this issue, we propose ViFP, a training-free framework for visual FP detection and correction. ViFP identifies FPs through consistency analysis between direct reasoning and structured multi-step reasoning, uncovering self-inconsistent reasoning without relying on superior supervisory models. Based on detected FPs, ViFP adaptively optimizes reasoning paths by refining question types and reasoning chain templates. Furthermore, we introduce a novel reliability metric, Value of Correction (VoC), which quantitatively measures the benefit of FP correction by jointly considering accuracy improvement, true negative rate, and FP reduction. VoC provides an interpretable indicator of reasoning reliability beyond conventional accuracy metrics. ViFP is designed as a cloud/edge-side reasoning reliability service for IoT systems and is compatible with leading closed-source VLMs. Experiments on A-OKVQA, OK-VQA, and FVQA demonstrate that ViFP significantly improves reasoning accuracy and reduces FPs, achieving up to 5.4% accuracy gain and surpassing previous state-of-the-art performance.

Large language models (LLMs) exhibit impressive performance across diverse tasks but often struggle to accurately gauge their knowledge boundaries, leading to confident yet incorrect responses.This paper explores leveraging LLMs' internal states to enhance their perception of knowledge boundaries from efficiency and risk perspectives.We investigate whether LLMs can estimate their confidence using internal states before response generation, potentially saving computational resources.Our experiments on datasets like Natural Questions, HotpotQA, and MMLU reveal that LLMs demonstrate significant pre-generation perception, which is further refined post-generation, with perception gaps remaining stable across varying conditions.To mitigate risks in critical domains, we introduce Consistency-based Confidence Calibration (C 3 ), which assesses confidence consistency through question reformulation.C 3 significantly improves LLMs' ability to recognize their knowledge gaps, enhancing the unknown perception rate by 5.6% on NQ and 4.9% on HotpotQA.Our findings suggest that pre-generation confidence estimation can optimize efficiency, while C 3 effectively controls output risks, advancing the reliability of LLMs in practical applications 1 .

Super-resolution reconstruction technology has important scientific significance and application value in the field of image processing by performing image restoration processing on one or more low-resolution images to improve image spatial resolution. Based on the SCSR algorithm and VDSR network, in order to further improve the image reconstruction quality, an image super-resolution reconstruction algorithm combined with multi-residual network and multi-feature SCSR(MRMFSCSR) is proposed. Firstly, at the sparse reconstruction stage, according to the characteristics of image blocks, our algorithm extracts the contour features of non-flat blocks by NSCT transform, extracts the texture features of flat blocks by Gabor transform, then obtains the reconstructed high-resolution (HR) images by using sparse models. Secondly, according to improve the VDSR deep network and introduce the feature fusion idea, the multi-residual network structure (MR) is designed. The reconstructed HR image obtained by the sparse reconstruction stage is used as the input of the MR network structure to optimize the high-frequency detail residual information. Finally, we can obtain a higher quality super-resolution image compared with the SCSR algorithm and the VDSR algorithm.