PolygoNet: Leveraging Simplified Polygonal Representation for Effective Image Classification

2025-04-01

AI Summary

To address the high computational cost, susceptibility to overfitting, and challenges in edge deployment of image classification models, this paper proposes a lightweight modeling approach based on differentiable polygonal geometric representation. Specifically, input images are first converted into dominant point sequences via the Douglas–Peucker algorithm and contour extraction; these sequences are then mapped—through an end-to-end differentiable geometric encoder—into structured geometric features that replace redundant pixel-based inputs. Crucially, we introduce the first integration of a lightweight polygonal representation directly into CNN backbones, enabling explicit geometric priors for feature compression, noise suppression, and implicit regularization. Evaluated on multiple benchmark datasets, our method achieves state-of-the-art (SOTA) accuracy under full-image evaluation, accelerates training by 3.2×, reduces model parameters by 87%, and successfully deploys on resource-constrained edge devices such as the Raspberry Pi.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Learning on the Edge & Model Compression
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Deep learning models have achieved significant success in various image related tasks. However, they often encounter challenges related to computational complexity and overfitting. In this paper, we propose an efficient approach that leverages polygonal representations of images using dominant points or contour coordinates. By transforming input images into these compact forms, our method significantly reduces computational requirements, accelerates training, and conserves resources making it suitable for real time and resource constrained applications. These representations inherently capture essential image features while filtering noise, providing a natural regularization effect that mitigates overfitting. The resulting lightweight models achieve performance comparable to state of the art methods using full resolution images while enabling deployment on edge devices. Extensive experiments on benchmark datasets validate the effectiveness of our approach in reducing complexity, improving generalization, and facilitating edge computing applications. This work demonstrates the potential of polygonal representations in advancing efficient and scalable deep learning solutions for real world scenarios. The code for the experiments of the paper is provided in https://github.com/salimkhazem/PolygoNet.

Problem

• Reducing computational complexity in image classification models
• Mitigating overfitting through polygonal representation regularization
• Enabling efficient deep learning on resource-constrained edge devices

Innovation

• Uses polygonal representations for image classification
• Reduces computational complexity and accelerates training
• Enables deployment on resource-constrained edge devices