Hierarchical Feature Fusion Triple Network for Change Detection With Bitemporal Remote Sensing Images | IEEE Journals & Magazine | IEEE Xplore

Hierarchical Feature Fusion Triple Network for Change Detection With Bitemporal Remote Sensing Images


Abstract:

Achieving land cover change detection (LCCD) through remotely sensed images (RSIs) is important in the observation of the changes on the Earth’s surface. In such detectio...Show More

Abstract:

Achieving land cover change detection (LCCD) through remotely sensed images (RSIs) is important in the observation of the changes on the Earth’s surface. In such detection, spectral-reflectance noise and the uncertainty of the imaging external conditions for the bitemporal RSIs usually cause some salt-and-pepper noisy pixels in the results and reduce the change detection accuracy. In this article, a hierarchical feature-fusion triple network (HFTN) is proposed to improve the performance of LCCD with RSIs. Overall, the proposed HFTN aims to learn representative features to improve change detection performance via two feature learning enhancement strategies and a hierarchical feature-fusion mechanism. First, an image feature difference model is proposed to generate the input feature for the middle branch and guide the learning performance. Second, a progressive denoising module (PDM) is proposed and applied to each temporal image to reduce the noise before feeding the features into the backbone of the proposed HFTN. Finally, a hierarchical feature-fusion module (HFFM) is proposed to fuse the learned deep feature for generating a change-magnitude image. Additionally, multiscale convolution, cross-scale fusion, and a shared weight are adopted in the backbone of the proposed HFTN to further enhance the feature learning performance. Compared with eight state-of-the-art methods, experimental results verified the feasibility and superiority of the proposed HFTN for LCCD with RSIs. For example, the proposed HFTN achieved improvement rates of approximately 0.43%–11.83% for overall accuracy (OA) and 0.11%–4.81% for false alarms (FAs) across six pairs of real RSIs. The code can be available at https://github.com/ImgSciGroup/HFTN-NET.git.
Article Sequence Number: 4401512
Date of Publication: 06 January 2025

ISSN Information:

Funding Agency:


I. Introduction

Land cover change detection (LCCD) with remotely sensed images (RSIs) is an important technology for observing Earth’s surface changes [1], [2], [3]. Owing to the advantages of remotely sensed technology in effectively capturing large-scale land cover [4], [5], [6], LCCD with RSIs has been widely used in land-use motoring [7], geological disaster detection [8], [9], natural resource management [10], and environmental monitoring [11]. However, LCCD usually refers to more than two or three temporal RSIs in a detection task, and some challenges are usually encountered in practical applications, as presented below.

Spectral noise is inevitable in an imaging process for RSIs [12]. The imaging progress depends on two aspects: the platform and the sensor. Intuitively, pre-processing operations, such as geometry and reflectance correction, are required for a raw RSI acquired with a remotely sensed platform (such as a satellite or an aerial plane). This operation may smoothen some noise or bring some noise to a processed image [13], [14]. Additionally, remote sensors belong to optical physical equipment, which produces some noise in the imaging process. Moreover, the noise from the equipment is inevitable.

Pseudo-change is observed as noise in a change detection map when using bitemporal RSIs [15]. The bitemporal RSIs used for LCCD cover the same geography but occur on different dates. External imaging conditions, such as atmospheric moisture, phenology difference, and sun height, cause pseudo-changes in an LCCD task [16], [17]. Additionally, mixing pixels may also occur in RSIs and result in accused pseudo-changes [18]. These pseudo-changes may be considered as noise in a detection map and reduce the detection accuracies.

Ground targets on the Earth’s surface with various shapes and sizes challenge the utilization of spatial-contextual features. Many studies have demonstrated that utilizing spatial-contextual features can improve the application performance with RSIs [19]. The ground targets in an image scene with different shapes and sizes result in changed and unchanged areas between bitemporal RSIs, also with varied shapes and sizes. Utilizing spatial-contextual features for LCCD with RSIs is a challenge.

References

References is not available for this document.