Source-linked AI summary
Face Mask Detection using Transfer Learning of InceptionV3
G. Jignesh Chowdary, Narinder Singh Punn, Sanjay Kumar Sonbhadra, Sonali Agarwal
TL;DR
COVID-19 created a need to monitor mask wearing in crowded public places, where manual monitoring is difficult. The paper proposes fine-tuning InceptionV3 with augmented SMFD training data for automated mask detection, and reports higher accuracy than compared models.
Problem
Manual monitoring of mask wearing in crowded public places is difficult, while mask detection research is described as limited.
Method
The paper fine-tunes a pre-trained InceptionV3 model for mask classification and uses image augmentation to increase training-data diversity.
Results
The proposed transfer-learning model achieved higher accuracy than the compared machine-learning and deep-learning models under the same environment.
Takeaways & Limitations
The paper develops an automated system that classifies people who are not wearing masks using transfer learning of InceptionV3.
Abstract
from arXiv · showhide
The world is facing a huge health crisis due to the rapid transmission of coronavirus (COVID-19). Several guidelines were issued by the World Health Organization (WHO) for protection against the spread of coronavirus. According to WHO, the most effective preventive measure against COVID-19 is wearing a mask in public places and crowded areas. It is very difficult to monitor people manually in these areas. In this paper, a transfer learning model is proposed to automate the process of identifying the people who are not wearing mask. The proposed model is built by fine-tuning the pre-trained state-of-the-art deep learning model, InceptionV3. The proposed model is trained and tested on the Simulated Masked Face Dataset (SMFD). Image augmentation technique is adopted to address the limited availability of data for better training and testing of the model. The model outperformed the other recently proposed approaches by achieving an accuracy of 99.9% during training and 100% during testing.
1 Introduction
The paper frames mask wearing as an important measure against COVID-19 transmission but notes that manually monitoring crowded public places is difficult. It therefore proposes an AI-based model to detect people not wearing masks.
- COVID-19 transmission is concentrated in crowded areas, where mask wearing is presented as a preventive measure.
- AI, machine learning, and deep learning are identified as tools for monitoring and addressing COVID-19-related challenges.
- The work develops a deep learning model using InceptionV3 transfer learning to detect people not wearing masks in public places.The model is intended for integration with surveillance cameras, while image augmentation increases training-data diversity.
2 Literature Review
Prior work addresses face recognition, reconstruction, and mask-related detection, but the paper identifies mask detection research as limited and proposes an InceptionV3 transfer-learning approach for further improvement.
- Earlier studies applied SRCNet, YOLOv3, GANs, and VGG16 to face-mask, face-reconstruction, operating-room, and facial-emotion tasks.Reported accuracies include 98.7% for SRCNet, 93.9% for YOLOv3, 95% for operating-room mask detection, and 88% for VGG16 emotion recognition.
- Mask detection research is described as limited, with further improvement needed in existing methods.
- The paper responds by proposing a transfer-learning approach that uses a trained InceptionV3 model for face-mask recognition.
3 Dataset Description
The study uses the Simulated Masked Face Dataset, containing balanced masked and unmasked facial images, with separate training and testing subsets.
- The SMFD contains 1,570 images: 785 simulated masked facial images and 785 unmasked facial images.
- The dataset is divided into 1,099 training images and 470 testing images.
- Sample dataset images are presented in Fig. 1.
4 Proposed Methodology
The methodology addresses limited data through image augmentation and performs mask classification with a fine-tuned InceptionV3 transfer-learning model.
- 4 Proposed Methodology: The methodology has two phases: augmenting training data to address limited samples, then detecting masks with InceptionV3 transfer learning.
- 4.1 Image augmentation: Image augmentation modifies training images using eight operations to increase dataset size and diversity.The generated images are rescaled to 224 x 224 pixels and converted to single-channel greyscale.
- 4.1 Image augmentation: Fig. 2 illustrates augmented training images, while Fig. 3 presents a schematic of the proposed methodology.
- 4.2 Transfer Learning: Transfer learning reuses trained neural-network weights to improve performance when training on a small dataset.The paper describes InceptionV3 as a 48-layer convolutional neural network developed by Google.
- 4.2 Transfer Learning: The proposed model removes InceptionV3’s last layer and adds pooling, flattening, dense, dropout, and two-neuron softmax layers.It is trained for 80 epochs with 42 steps per epoch.
5 Results
The study evaluates an InceptionV3 transfer-learning model for face-mask detection using multiple performance metrics and comparisons with other models. Results include confusion-matrix analysis, training/testing comparisons, and sample detection outputs.
- The model is evaluated using Accuracy, Precision, Sensitivity, Specificity, Intersection over Union, and Matthews Correlation Coefficient.
- Two confusion matrices are constructed to evaluate model performance during training and testing.
- The proposed model achieves higher accuracy than decision tree, support vector machine, MobileNet, MobileNetV2, Xception, VGG16, and VGG19 models under the same environment.
- Sample outputs show face bounding boxes, mask-status colors, and confidence scores.
6 Conclusion
The paper presents an InceptionV3 transfer-learning model with image augmentation to detect people not wearing masks in public and crowded places. On the SMFD dataset, it reports very high training and testing accuracy and specificity, while noting that larger data volumes could improve the work.
- The proposed model detects people not wearing masks using InceptionV3 transfer learning and image augmentation.
- 99.92% accuracy and 99.9% specificity are achieved during training on the SMFD dataset.
- 100% accuracy and 100% specificity are achieved during testing on the SMFD dataset.
- The work can be further improved by employing larger volumes of data.