arXiv:2511.12346v2 Announce Type: replace 
Abstract: Hyperspectral image (HSI) classification faces critical challenges, including high spectral dimensionality, complex spectral-spatial correlations, and limited training samples with severe class imbalance. While CNNs excel at local feature extraction and transformers capture long-range dependencies, their isolated application yields suboptimal results due to quadratic complexity and insufficient inductive biases. We propose CLAReSNet (Convolutional Latent Attention Residual Spectral Network), a hybrid architecture that integrates multi-scale convolutional extraction with transformer-style attention via an adaptive latent bottleneck. The model employs a multi-scale convolutional stem with deep residual blocks and an enhanced Convolutional Block Attention Module for hierarchical spatial features, followed by spectral encoder layers combining bidirectional RNNs (LSTM/GRU) with Multi-Scale Spectral Latent Attention (MSLA). MSLA reduces complexity from $\mathcal{O}(T^2D)$ to $\mathcal{O}(T\log(T)D)$ by adaptive latent token allocation (8-64 tokens) that scales logarithmically with the sequence length. Hierarchical cross-attention fusion dynamically aggregates multi-level representations for robust classification. Experiments conducted on the Indian Pines and Salinas datasets show state-of-the-art performance, achieving overall accuracies of 99.71% and 99.96%, significantly surpassing HybridSN, SSRN, and SpectralFormer. The learned embeddings exhibit superior inter-class separability and compact intra-class clustering, validating CLAReSNet's effectiveness under severe class imbalance.

تم تقديم CLAReSNet، وهي بنية هجينة جديدة لتصنيف الصور الطيفية، لمواجهة التحديات مثل الأبعاد الطيفية العالية وعدم توازن الفئات. يجمع هذا النموذج بين استخراج التلافيف متعددة المقاييس مع الانتباه على طراز المحولات من خلال عنق زجاجة كامن قابل للتكيف، مما يعزز عملية التصنيف.

CLAReSNet, una nueva arquitectura híbrida para la clasificación de imágenes hiperespectrales, ha sido presentada para abordar desafíos como la alta dimensionalidad espectral y el desequilibrio de clases. Este modelo combina la extracción convolucional a múltiples escalas con atención de estilo transformador a través de un cuello de botella latente adaptativo, mejorando así el proceso de clasificación.

CLAReSNet, une nouvelle architecture hybride pour la classification des images hyperspectrales, a été introduite pour relever des défis tels que la haute dimensionnalité spectrale et le déséquilibre des classes. Ce modèle combine l'extraction convolutive multi-échelle avec une attention de type transformateur via un goulot d'étranglement latent adaptatif, améliorant ainsi le processus de classification.

CLAReSNet, a novel hybrid architecture for hyperspectral image classification, has been introduced to tackle challenges such as high spectral dimensionality and class imbalance. This model combines multi-scale convolutional extraction with transformer-style attention through an adaptive latent bottleneck, enhancing the classification process.

CLAReSNet: When Convolution Meets Latent Attention for Hyperspectral Image Classification

Was this article worth reading? Share it

LucidQuery AI

Attentive AI

The Visualizer

ClousTech

Lenso.ai

LexiStock AI

Ready to build your own newsroom?