Scholarlink

<p>The COVID-19 pandemic highlighted the importance of face mask usage in public health safety, giving way to automated face mask detection, especially for real-time surveillance. This paper presents a comprehensive approach for binary face mask classification using deep learning techniques. We propose a Stacked Neural Ensembles combining different deep learning techniques (Convolutional Neural Network, MobileNetV2, and ResNet50), with a shallow neural network as the meta-learner. Further, we evaluate multiple conventional machine learning algorithms such as Support Vector Machines, Naive Bayes, k-Nearest Neighbors, and Decision Trees, for classifying images as “With Mask” or “Without Mask.” The models were implemented on two publicly available datasets: Real-World Masked Face Dataset (RMFD) and Kaggle Face Mask 12k dataset (FM12k). The performance was measured through accuracy-loss curves and classification reports. The experimental results revealed that the proposed Stacked Neural Ensemble model with a shallow neural network as the meta-learner showed the effectiveness of model fusion by achieving better results than the individual models with an accuracy of 0.995 on FM12k and 0.99 on RMFD. These findings show that ensemble learning can increase generalization and reliability in important health applications.</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

Enhancing face mask detection performance using stacked neural ensembles in mask fusion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月31日

查看全文
  • Rakhee Kumari,
  • Pallavi Pallavi,
  • Praneet Saurabh

摘要

The COVID-19 pandemic highlighted the importance of face mask usage in public health safety, giving way to automated face mask detection, especially for real-time surveillance. This paper presents a comprehensive approach for binary face mask classification using deep learning techniques. We propose a Stacked Neural Ensembles combining different deep learning techniques (Convolutional Neural Network, MobileNetV2, and ResNet50), with a shallow neural network as the meta-learner. Further, we evaluate multiple conventional machine learning algorithms such as Support Vector Machines, Naive Bayes, k-Nearest Neighbors, and Decision Trees, for classifying images as “With Mask” or “Without Mask.” The models were implemented on two publicly available datasets: Real-World Masked Face Dataset (RMFD) and Kaggle Face Mask 12k dataset (FM12k). The performance was measured through accuracy-loss curves and classification reports. The experimental results revealed that the proposed Stacked Neural Ensemble model with a shallow neural network as the meta-learner showed the effectiveness of model fusion by achieving better results than the individual models with an accuracy of 0.995 on FM12k and 0.99 on RMFD. These findings show that ensemble learning can increase generalization and reliability in important health applications.