Hierarchical Graph Feature Enhancement with Adaptive Frequency Modulation for Visual Recognition

2025-08-15

AI Summary

CNNs’ inherent grid-based architecture struggles to capture complex image topology and non-local semantic relationships. To address this, we propose the Hierarchical Graph Feature Enhancement (HGFE) framework, which constructs a dual-level graph structure—local window graphs and global supernode graphs—to jointly encode local geometric constraints and global semantic correlations. We further introduce an adaptive frequency modulation module that dynamically regulates high- and low-frequency information propagation, effectively mitigating over-smoothing while preserving edge and texture details. All components—including intra-window graph convolution, inter-window supernode interaction, and frequency modulation—are lightweight, end-to-end trainable, and modularly integrable into mainstream CNN backbones without architectural modification. Extensive experiments on diverse benchmarks—including CIFAR-100 (classification), PASCAL VOC (detection), and VisDrone, CrackSeg, and CarParts (segmentation)—demonstrate consistent and significant performance gains across all tasks.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Convolutional neural networks (CNNs) have demonstrated strong performance in visual recognition tasks, but their inherent reliance on regular grid structures limits their capacity to model complex topological relationships and non-local semantics within images. To address this limita tion, we propose the hierarchical graph feature enhancement (HGFE), a novel framework that integrates graph-based rea soning into CNNs to enhance both structural awareness and feature representation. HGFE builds two complementary levels of graph structures: intra-window graph convolution to cap ture local spatial dependencies and inter-window supernode interactions to model global semantic relationships. Moreover, we introduce an adaptive frequency modulation module that dynamically balances low-frequency and high-frequency signal propagation, preserving critical edge and texture information while mitigating over-smoothing. The proposed HGFE module is lightweight, end-to-end trainable, and can be seamlessly integrated into standard CNN backbone networks. Extensive experiments on CIFAR-100 (classification), PASCAL VOC, and VisDrone (detection), as well as CrackSeg and CarParts (segmentation), validated the effectiveness of the HGFE in improving structural representation and enhancing overall recognition performance.

Problem

• Enhancing CNNs' ability to model complex topological relationships in images
• Addressing limitations in capturing non-local semantics and spatial dependencies
• Mitigating over-smoothing while preserving critical edge and texture information

Innovation

• Hierarchical graph feature enhancement for structural awareness
• Adaptive frequency modulation balancing signal propagation
• Intra-window and inter-window graph convolution integration