Presented at the 6th International Symposium on Innovative Approaches in Smart Technologies (ISAS-WINTER-2022), Online, Turkey, Dec 08, 2022

SETSCI Conference Proceedings, 2022, 14, Page (s): 34-37 , https://doi.org/10.36287/setsci.5.2.007

Published Date: 22 December 2022    | 1476     15

In the last decade, deep learning in general and Convolutional Neural Networks (CNN) have achieved a human level and outstanding performance in almost all computer-based systems such as classification(text, images, and videos). Building an efficient CNN architecture and finding the most suitable layers requires time. In addition, tuning and setting CNN parameters such as the number of filters, size of filters, Dense rate, Dropout value, etc., has an essential effect on the overall performance of the CNN model.

This paper investigated the effects of CNN hyperparameter optimizations using Random Search. To achieve our goals, we selected to investigate the hyperparameter optimizations of MobileNetV2 and VGG19 models when used for image classification. Here, we have combined four mask detection datasets of approximately 6K, 12K, 4k, and 4k  images into one dataset.

Overall, the results of the experimental works showed that using the hyperparameter optimization technique improved the overall performance of both MobileNetV2 and VGG19 models. In addition, VGG19 outperformed MobileNetV2 across Accuracy, Precision, recall, and F1 score evaluation matrixes.

Keywords - CNN, Deep learning, image classification, random search, hyperparameter optimization, Neural Architecture Search

[1] Liashchynskyi, P., & Liashchynskyi, P. (2019). Grid search, random search, genetic algorithm: a big comparison for NAS. arXiv preprint arXiv:1912.06059.

[2] Li, L., & Talwalkar, A. (2020, August). Random search and reproducibility for neural architecture search. In Uncertainty in artificial intelligence (pp. 367-377). PMLR.

[3] Mellor, J., Turner, J., Storkey, A., & Crowley, E. J. (2021, July). Neural architecture search without training. In International Conference on Machine Learning (pp. 7588-7598). PMLR.

[4] VIJAY KUMAR, (2021) Face Mask Detection, [Online]. Available: https://www.kaggle.com/datasets/vijaykumar1799/face-mask-detection.

[5] ASHISH JANGRA, (2021) Face Mask Detection~12K Images Dataset, [Online]. Available: https://www.kaggle.com/datasets/ashishjangra27 /face-mask-12k-images-dataset

[6] Real World Fasked Face Recognition Dataset (RMFRD), (2021) [Online]. Available: https://www.kaggle.com/datasets/ muhammeddalkran/masked-facerecognition.

[7] Balaji S, (2021) Face-Mask-Detection, [Online]. Available: https://github.com/balajisrinivas/Face-Mask-Detection/tree/master /dataset/ with_mask

[8] Teboulbi S, Messaoud S, Hajjaji MA, Mtibaa A., “Face Mask Classification Based on Deep Learning Framework,” InAdvanced Practical Approaches to Web Mining Techniques and Application, pp. 175-188, IGI Global, 2022.

[9] Sandler M, Howard A, Zhu M, Zhmoginov A, Chen LC., “Mobilenetv2: Inverted residuals and linear bottlenecks,” InProceedings of the IEEE conference on computer vision and pattern recognition, pp. 4510-4520, 2018.