MSFF: A Multi-Scale Feature Fusion Convolutional Neural Network for Hyperspectral Image Classification

• DOI: 10.3390/electronics14040797
• Publication Date: 2025-02-18T14:45:23Z
• Authors: Gu Gong; Xiaopeng Wang; Jiahua Zhang; Xiaodi Shang; Zhicheng Pan; Zhiyuan Li; Junshi Zhang
• Keywords: Feature (linguistics)

In contrast to conventional remote sensing images, hyperspectral images are characterized by a greater number of spectral bands and exceptionally high resolution. The richness both spatial information facilitates the precise classification various objects within establishing imaging as indispensable for applications. However, labor-intensive time-consuming process labeling results in limited labeled samples, while challenges like similarity between different variation same object further complicate development algorithms. Therefore, efficiently exploiting is crucial accomplishing task. To address these challenges, this paper presents multi-scale feature fusion convolutional neural network (MSFF). introduces dual branch extraction module utilizing 3D depthwise separable convolution joint extraction, refined an attention-based-on-central-pixels (ACP) mechanism. Additionally, spectral–spatial attention (SSJA) designed interactively explore latent dependency through use multilayer perceptron global pooling operations. Finally, (FF) adaptive (AMSFE) incorporated enable comprehensive mining information. Experimental demonstrate that proposed method performs well on IP, PU, YRE datasets, delivering superior compared other methods underscoring potential advantages MSFF classification.