Scholarlink

<p>This study presents an advanced deep learning framework for vehicle damage detection, featuring an innovative feature selection method, DeepChiOptimizer. By automating deep feature selection and classifier configuration through a dynamic grid search process, DeepChiOptimizer minimizes classification errors and optimizes performance. Unlike traditional approaches such as Least Absolute Shrinkage and Selection Operator (LASSO), Principal Component Analysis (PCA), and Recursive Feature Elimination (RFE), this method adaptively selects the most informative deep features while concurrently determining optimal classifier configurations. Eliminating the trial-and-error process for selecting abstract features from deep networks ensures the precise identification of feature-classifier combinations with superior accuracy. The framework’s effectiveness is validated across datasets of varying complexity, utilizing hybrid combinations of deep learning architectures. High-level features are extracted from multiple pre-trained Convolutional Neural Network (CNN) models, including VGG16, DenseNet121, EfficientNetB0, ResNet50, and MobileNetV2, and concatenated into a single vector to maximize feature representation. The effectiveness of the proposed framework is validated using a hybrid model that combines VGG16, DenseNet121, DeepChiOptimizer, and Support Vector Machine (SVM), achieving an accuracy of 98.33%. This enhances speed and accuracy in insurance damage assessments while reducing manual work, streamlining claims, minimizing costs, and improving decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DeepChiOptimizer for Automated Vehicle Damage Detection Using Hybrid Deep Learning and Feature Selection

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Öznur Özaltın

摘要

This study presents an advanced deep learning framework for vehicle damage detection, featuring an innovative feature selection method, DeepChiOptimizer. By automating deep feature selection and classifier configuration through a dynamic grid search process, DeepChiOptimizer minimizes classification errors and optimizes performance. Unlike traditional approaches such as Least Absolute Shrinkage and Selection Operator (LASSO), Principal Component Analysis (PCA), and Recursive Feature Elimination (RFE), this method adaptively selects the most informative deep features while concurrently determining optimal classifier configurations. Eliminating the trial-and-error process for selecting abstract features from deep networks ensures the precise identification of feature-classifier combinations with superior accuracy. The framework’s effectiveness is validated across datasets of varying complexity, utilizing hybrid combinations of deep learning architectures. High-level features are extracted from multiple pre-trained Convolutional Neural Network (CNN) models, including VGG16, DenseNet121, EfficientNetB0, ResNet50, and MobileNetV2, and concatenated into a single vector to maximize feature representation. The effectiveness of the proposed framework is validated using a hybrid model that combines VGG16, DenseNet121, DeepChiOptimizer, and Support Vector Machine (SVM), achieving an accuracy of 98.33%. This enhances speed and accuracy in insurance damage assessments while reducing manual work, streamlining claims, minimizing costs, and improving decision-making.