Domain-Aware Lightweight Spectral-Grouped Convolutions for Hyperspectral Fish Freshness Classification

2026-08-12

AI Summary

This study addresses the challenge of accurately classifying fish freshness from hyperspectral images, which is hindered by spectral dominance, ordinal label structure, and severe sample scarcity that limit the effectiveness of conventional deep learning approaches. To overcome these limitations, this work proposes SGNet, a lightweight network that introduces domain-aware spectral grouping convolution to decouple spectral and spatial features. By integrating depthwise separable convolution with a dual channel–spatial attention mechanism, SGNet enhances discriminative capability with minimal computational overhead. Evaluated on a self-collected dataset of salmon samples stored over 16 days, SGNet achieves a classification accuracy of 97.8% and an average absolute error of 0.64 days using only 4.75 million parameters—reducing parameter count by 5–18× compared to ResNet-50 and Vision Transformer while maintaining high accuracy and real-time performance.

Abstract

Hyperspectral imaging (HSI) offers nondestructive assessment of fish freshness by detecting biochemical alterations across spectral bands. However, conventional deep learning approaches do not fully address the particular characteristics of HSI data, such as spectral dominance over spatial textures, ordinal label structure, and a small number of training samples. We propose SGNet (Spectral-Grouped Network), a lightweight architecture that separates spectral and spatial feature extraction using grouped convolutions and a depthwise spatial pathway. A dual attention mechanism that couples channel-wise squeeze-and-excitation with spatial gating adaptively highlights informative features. SGNet achieves 97.8% classification accuracy and 0.64 days mean absolute error (MAE) with just 4.75M parameters when tested on our newly developed 16-day refrigerator-stored salmon fillet dataset. Ablation studies validate the contribution of each component, while comparisons demonstrate a five- to eighteen-fold parameter reduction relative to ResNet-50 and Vision Transformers. Our findings indicate that domain-aware design supports precise, real-time freshness prediction for industrial implementation.

Problem

• hyperspectral imaging
• fish freshness classification
• spectral-spatial features
• ordinal labels
• limited training samples

Innovation

• spectral-grouped convolutions
• dual attention mechanism
• lightweight architecture
• hyperspectral image classification
• domain-aware design