Multi-Focus Image Fusion Based on Fractal Dimension and Parameter Adaptive Unit-Linking Dual-Channel PCNN in Curvelet Transform Domain

Fractal and Fractional, Journal Year: 2025, Volume and Issue: 9(3), P. 157 - 157

Published: March 3, 2025

Multi-focus image fusion is an important method for obtaining fully focused information. In this paper, a novel multi-focus based on fractal dimension (FD) and parameter adaptive unit-linking dual-channel pulse-coupled neural network (PAUDPCNN) in the curvelet transform (CVT) domain proposed. The source images are decomposed into low-frequency high-frequency sub-bands by CVT, respectively. FD PAUDPCNN models, along with consistency verification, employed to fuse sub-bands, average used sub-band, final fused generated inverse CVT. experimental results demonstrate that proposed shows superior performance Lytro, MFFW, MFI-WHU datasets.

Language: Английский

---

Multi-focus image fusion based on pulse coupled neural network and WSEML in DTCWT domain

Frontiers in Physics, Journal Year: 2025, Volume and Issue: 13

Published: April 2, 2025

The goal of multi-focus image fusion is to merge near-focus and far-focus images the same scene obtain an all-focus that accurately comprehensively represents focus information entire scene. current algorithms lead issues such as loss details edges, well local blurring in resulting images. To solve these problems, a novel method based on pulse coupled neural network (PCNN) weighted sum eight-neighborhood-based modified Laplacian (WSEML) dual-tree complex wavelet transform (DTCWT) domain proposed this paper. source are decomposed by DTCWT into low- high-frequency components, respectively; then average gradient (AG) motivate PCNN-based rule used process low-frequency WSEML-based components; we conducted simulation experiments public Lytro dataset, demonstrating superiority algorithm proposed.

Language: Английский

---

Robust Infrared–Visible Fusion Imaging with Decoupled Semantic Segmentation Network

Sensors, Journal Year: 2025, Volume and Issue: 25(9), P. 2646 - 2646

Published: April 22, 2025

The fusion of infrared and visible images provides complementary information from both modalities has been widely used in surveillance, military, other fields. However, most the available methods have only evaluated with subjective metrics visual quality fused images, which are often independent following relevant high-level tasks. Moreover, as a useful technique especially low-light scenarios, effect conditions on result not well-addressed yet. To address these challenges, decoupled semantic segmentation-driven image network is proposed this paper, connects downstream task to drive be optimized. Firstly, cross-modality transformer module designed learn rich hierarchical feature representations. Secondly, semantic-driven developed enhance key features prominent targets. Thirdly, weighted strategy adopted automatically adjust weights different modality features. This effectively merges thermal characteristics detailed images. Additionally, we design refined loss function that employs decoupling constrain pixel distributions produce more-natural evaluate robustness generalization method practical challenge applications, Maritime Infrared Visible (MIV) dataset created verified for maritime environmental perception, will made soon. experimental results public datasets practically collected MIV highlight notable strengths best-ranking among its counterparts. Of more importance, achieved over 96% target detection accuracy dominant high mAP@[50:95] value far surpasses all competitors.

Language: Английский

---

Leveraging Land Cover Priors for Isoprene Emission Super-Resolution

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(10), P. 1715 - 1715

Published: May 14, 2025

Satellite remote sensing plays a crucial role in monitoring Earth’s ecosystems, yet satellite-derived data often suffer from limited spatial resolution, restricting the availability of accurate and precise for atmospheric modeling climate research. Errors biases may also be introduced into applications due to use with insufficient temporal resolution. In this work, we propose deep learning-based Super-Resolution (SR) framework that leverages land cover information enhance accuracy Biogenic Volatile Organic Compound (BVOC) emissions, particular focus on isoprene. Our approach integrates priors as emission drivers, capturing patterns more effectively than traditional methods. We evaluate model’s performance across various conditions analyze statistical correlations between isoprene emissions key environmental such cropland tree data. Additionally, assess generalization capabilities our SR model by applying it unseen zones geographical regions. Experimental results demonstrate incorporating significantly improves accuracy, particularly heterogeneous landscapes. This study contributes chemistry providing cost-effective, data-driven refining BVOC maps. The proposed method enhances usability satellite-based data, supporting air quality forecasting, impact assessments, studies.

Language: Английский

---