Integrating Mathematical Models for Enhanced Object Detection: a Hybrid Deep Learning Approach

EasyChair Preprint 15596

Abstract

This paper investigates the mathematical principles underlying hybrid object detection models that combine Convolutional Neural Networks (CNNs) with Vision Transformers (ViTs). We present a comprehensive mathematical framework for feature extraction, attention mechanisms, and optimization techniques. By incorporating advanced regularization methods and custom loss functions, our goal is to enhance detection accuracy while minimizing computational costs. Notable contributions include mathematical formulations for attention-aware convolutional layers and a dynamic loss function designed to balance localization and classification errors effectively.

Keyphrases: Algorithms, CNN, ViT, deep learning

@booklet{EasyChair:15596,
  author    = {Mo Zhang and Kon Yang and Mo Chang and Michael Lornwood},
  title     = {Integrating Mathematical Models for Enhanced Object Detection: a Hybrid Deep Learning Approach},
  howpublished = {EasyChair Preprint 15596},
  year      = {EasyChair, 2024}}