.png){kind=icon}
返回
Enhancing deep learning image classification using data augmentation and genetic algorithm-based optimization
.png){kind=icon}
PRE
AI
DOI：10.1007/s13735-024-00345-5.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
In machine learning, data augmentation stands out as a potent strategy to overcome the constraints imposed by limited training data, as well as unbalanced and low-quality data, ultimately enhancing model accuracy. This work introduces a novel data augmentation technique rooted in optimization, seamlessly integrating a genetic algorithm. It unlocks a broad spectrum of possibilities for generating high-quality images and creating varied and distinctive images, thereby enriching the dataset. The proposed approach allows for identifying and selecting the most significant pixels within a given image, preserving vital information crucial for efficiently training deep learning models. Experimental evaluations were carried out on two datasets, namely Kaggle Cats versus Dogs and the Chest X-ray dataset. To underscore our proposed data augmentation technique's robustness and adaptability, we trained on different models: VGG16, VGG19, InceptionV3, EfficientNet-B0, and Vision Transformer. Our method consistently achieved peak accuracy while training these models on the specified datasets. Augmenting the Cat versus Dog dataset resulted in accuracies of 93.47%, 92.38%, 98.03%, 98.40%, and 78.61% for VGG16, VGG19, InceptionV3, EfficientNet-B0, and Vision Transformer, respectively. For the Chest X-ray dataset, the accuracies were 79.81%, 82.05%, 90.71%, 93.91% and 80.89% for VGG16, VGG19, InceptionV3, EfficientNet-B0 and Vision Transformer, respectively. For both datasets and each model, the proposed method consistently outperformed the accuracy achieved when considering the original dataset and randomly augmented datasets, underscoring the significant impact of our research.
Keyword:
Artificial intelligence
Image classification and analysis
Data augmentation
Genetic algorithm
Features selection
Optimisation
Deep learning
期刊
IF:
2.9
论文数:
279
被引数:
866
机构
引用论文
Data augmentation for univariate time series forecasting with neural networks.png)
PATTERN RECOGNITION
PATTERN RECOGNITIONIF7.6
PREAI
分享收藏Using Machine Learning Approach to Evaluate the PM2.5 Concentrations in China from 1998 to 2016
PREAI
分享收藏
Structural study of lanthanides(III) in aqueous nitrate and chloride solutions by EXAFS通过EXAFS对硝酸盐和氯化物水溶液中镧系元素 (III) 的结构研究
PREAI
分享收藏New oxypregnane glycosides from appetite suppressant herbal supplement Hoodia gordoniiSteroids
SteroidsIF0
PREAI
分享收藏
查看更多内容