Convolutional networks outperform attention models for ice sheet images
Why Convolution Still Matters: Evaluating Inductive Biases in Cryospheric Image Classification
Computer Vision and Pattern RecognitionMachine Learning
Summary
Classifying images of ice sheet surfaces is hard because many surface types look similar and some types are rare. The authors compared different computer models for identifying these surfaces, including older convolutional neural networks (CNNs), newer attention-based models, and hybrids. They found that traditional CNNs were generally more accurate and consistent, especially for surface textures, while hybrid models better detected rare ice types. This suggests simpler models with built-in assumptions about image patterns still work best for this kind of icy landscape data.
What this means in practice
- •For remote sensing teams: Improve ice sheet surface classification accuracy by choosing convolutional neural networks tuned for texture-rich cryosphere images.
- •For environmental monitoring squads: Deploy hybrid convolution-attention models to better detect rare and structurally distinct ice surface classes in satellite data.
Tested on one dataset.
Authors
Chhaya Kulkarni, Emam Hossain
Abstract
Recent advances in attention-based deep learning have motivated their adoption for remote sensing image classification; however, their benefits for cryospheric imagery, where surface states are dominated by fine-grained textures and class imbalance, remain unclear. In this work, we revisit a benchmark Greenland Ice Sheet image dataset, previously shown to favor convolutional neural networks (CNNs), to examine whether modern attention-based and hybrid architectures improve class-wise reliability. We conduct a controlled comparison between a classical CNN (AlexNet), a modern CNN (ConvNeXt-Tiny), a pure attention-based model (Swin-Tiny), and a hybrid convolution-attention model (CoAtNet-0) under identical training and evaluation protocols. Results show that AlexNet achieves the highest accuracy and the strongest balanced performance as measured by macro-averaged F1, while ConvNeXt-Tiny exhibits the highest macro-averaged AUC, indicating strong class separability but less consistent final decision quality. Class-wise analysis reveals that hybrid architectures improve recall for rare and structurally distinct surface classes, whereas convolutional models remain more reliable for texture-dominated categories. These findings highlight the importance of aligning architectural inductive bias with cryospheric data characteristics and suggest that increased model complexity does not necessarily translate to improved reliability for ice-sheet surface classification.