Batch Normalization을 결합한 CNN과 MNIST 데이터셋에서의 정확도 향상에 관한 연구
Improving Test Accuracy on the MNIST Dataset using a Simple CNN with Batch Normalization
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 논문은 MNIST 데이터셋을 활용한 손글씨 숫자 인식에서 합성곱 신경망(CNN)과 배치정규화(BN)를결합한 모델을 제안한다. LeCun et al.의 LeNet-5 모델의 성과를 뛰어넘는 것을 목표로 6계층 신경망구조를 설계하였다. 제안된 모델은 28x28 픽셀 이미지를 입력으로 받아 합성곱, 맥스 풀링, 완전연결계층을거쳐 처리하며, 특히 배치정규화계층을 도입하여 학습 안정성과 성능을 향상시켰다. 실험에서는 60,000개의훈련 이미지와 10,000개의 테스트 이미지를 사용하였으며, Momentum 최적화 알고리즘을 적용하였다. 모델 구성에서는 30개의 필터, 필터 사이즈 5x5, 패딩 0, 스트라이드 1을 사용하였고, ReLU 활성화 함수를채택하였다. 훈련 과정에서는 미니배치 사이즈 100, 총 20 에포크, 학습률 0.1로 설정하였다. 결과적으로제안된 모델은 99.22%의 테스트 정확도를 달성하여 LeNet-5의 99.05%를 상회하였으며, F1-score 0.9919를기록하여 모델의 성능을 입증하였다. 또한, 본 논문에서 제안한 6계층 모델은 LeCun et al.의 LeNet-5(7계층모델)와 Ji, Chun and Kim(10계층 모델)이 제안한 모델보다 더 단순한 구조로 모델의 효율성을 강조하였다. 본 연구의 결과는 AI 비전 검사기 등 실제 산업 응용에서 활용 가능성을 보여주며, 특히 스마트팩토리에서부품의 불량 상태를 판별하는 데 효과적으로 적용될 수 있을 것으로 기대된다.
Abstract
In this paper, we proposes a Convolutional Neural Networks(CNN) equipped with Batch Normalization(BN) for handwritten digit recognition training the MNIST dataset. Aiming to surpass the performance of LeNet-5 by LeCun et al., a 6-layer neural network was designed. The proposed model processes 28x28 pixel images through convolution, Max Pooling, and Fully connected layers, with the batch normalization to improve learning stability and performance. The experiment utilized 60,000 training images and 10,000 test images, applying the Momentum optimization algorithm. The model configuration used 30 filters with a 5x5 filter size, padding 0, stride 1, and ReLU as activation function. The training process was set with a mini-batch size of 100, 20 epochs in total, and a learning rate of 0.1. As a result, the proposed model achieved a test accuracy of 99.22%, surpassing LeNet-5's 99.05%, and recorded an F1-score of 0.9919, demonstrating the model's performance. Moreover, the 6-layer model proposed in this paper emphasizes model efficiency with a simpler structure compared to LeCun et al.'s LeNet-5 (7-layer model) and the model proposed by Ji, Chun and Kim (10-layer model). The results of this study show potential for application in real industrial applications such as AI vision inspection systems. It is expected to be effectively applied in smart factories, particularly in determining the defective status of parts.
참고문헌 (12)
- S. B. Lee, "Applying Batch Normalization to the MNIST Dataset", Quantitative Bio-Science, 42(2), pp.133~137 (2023)[DOI]
- M. R. Han, "A Study on the AI Based MNIST Handwriting Recognition", Jounal of The Korea Society of Information Technology Policy & Management, 11(6), pp.1509~1517 (2019)
- Y. Lecun, "Gradient-based learning applied to document recognition", Proceedings of the IEEE, 86(11), pp.2278~2324 (1998)
- S. Ioffe, "Batch normalization : Accelerating deep network training by reducing internal covariate shift", Proceedings of the International Conference on Machine Learning, 37, pp.448~456 (2015)
- M. G. Ji, "An Improved Image Classification Using Batch Normalization and CNN", Journal of Internet Computing and Services, 19(3), pp.35~42 (2018)[DOI]
- A. Krizhevsky, "ImageNet classification with deep convolutional neural networks", Communications of the ACM, 60(6), pp.84~90 (2017)[DOI]
- S. H. Park, "CNN-Based Linux Malware Detection Using XAI", Journal of Korea Multimedia Society, 27(4), pp.514~523 (2024)[DOI]
- K. S. Jang, "Rotation-Invariant Handwritten Numeral Recognition Using Convolution Neural Network", Journal of the Korea Institute of Information and Communication Engineering, 27(12), pp.1501~1507 (2023)
- K. He, "Deep Residual Learning for Image Recognition", IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2016)
- S. Kim, "Development and Application of an Artificial Intelligence Education Program for Elementary School Students : Focusing on CNN", The Korean Association Of Computer Education, 27(3), pp.43~54 (2024)[DOI]
- K. Saitoh, "Deep Learning from Scratch" (2017), Hanbit Media, Inc
- J. S. Kim, "A Study on the Improvement of MNIST Numbers Learning Speed by Changing the Weight of the Feature Map", Journal of Korean Institute of Information Technology, 18(3), pp.109~116 (2020)[DOI]