TITLE:
An Attention-Based Improved Laplacian Pyramid Network for Remote Sensing Image Pansharpening
AUTHORS:
Jiazheng Zhou
KEYWORDS:
Pansharpening, Laplacian Pyramid, Squeeze and Excitation Attention, Remote Sensing Images, Convolutional Neural Networks, Spectral Angle Mapper, Pan Chromatic, Multispectral
JOURNAL NAME:
Open Journal of Applied Sciences,
Vol.16 No.5,
May
27,
2026
ABSTRACT: With the rapid development of Earth observation technology, remote sening images have been widely used in land resource investigation, environmental monitoring, precision agriculture, urban planning, and disaster assessment. However, due to sensor imaging mechanisms, signal to noise ratio limitations, and data transmission constraints, it is difficult for a single satellite sensor to obtain images with both high spatial resolution and rich spectral information. Panchromatic (PAN) images usually provide spatial details and clear texture structures, while multispectral (MS) images contain richer spectral information but have lower spatial resolution. Pansharpening aims to fuse PAN and MS images to generate high resolution multispectral images, which is important for interpretation and analysis tasks. Existing pansharpening methods mainly include traditional and deep learning based approaches. Traditional methods, such as component substitution, multi resolution analysis, and model based optimization, are simple and efficient, but they often suffer from spectral distortion or insufficient spatial detail recovery. In recent years, convolutional neural networks have shown strong representation ability in pansharpening tasks. Nevertheless, many existing deep learning based methods still have limitations in multi scale feature utilization and spectral fidelity preservation. Spatial details are distributed at different resolution levels, and single scale feature extraction may fail to recover high frequency textures. In addition, different spectral channels contribute unequally to fusion, while pixel wise losses cannot directly constrain spectral direction consistency. To address these problems, this paper proposes an attention based improved Laplacian pyramid network for remote sensing image pansharpening. The proposed method uses an MTF based Laplacian pyramid to decompose PAN and upsampled MS images into multi scale components. An SEFCNN adaptive fusion module is introduced at each pyramid level to enhance important spatial spectral features through channel attention. Furthermore, a hybrid loss function combining multi scale reconstruction loss and spectral angle mapper loss is designed to improve spectral preservation. Experiments on QuickBird and WorldView-3 datasets are conducted to verify the effectiveness of the proposed method.