Exploring the Effectiveness of Optimized Convolutional Neural Network in Transfer Learning for Image Classification: A Practical Approach
Exploring the Effectiveness of Optimized Convolutional Neural Network in Transfer Learning for Image Classification: A Practical Approach
复制标题
探索优化卷积神经网络在图像分类迁移学习中的有效性：一种实用方法
DOI：
10.1109/incacct57535.2023.10141701
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Anupam Baliyan
中科院分区：
文献类型：
--
作者：
Srinivasa Rao Burri;S. Ahuja;Abhishek Kumar;Anupam Baliyan
迁移学习是一种流行的深度学习技术，涉及在新数据集上微调预训练的CNN模型，以提高准确性和速度。本文研究了迁移学习技术在使用CNN的图像分类任务中的有效性。本文回顾了最近关于迁移学习技术的研究，包括它们在医学图像分析应用中的应用，如COVID-19检测和阿尔茨海默病分类。该研究讨论了ImageNet数据集作为预训练CNN模型的基准，并提出了一种优化的CNN模型，该模型使用各种优化技术来提高性能和效率。本文还包括各种图像分类技术的比较表，包括CNN，RNN，SVM，RF和优化的CNN，优化的CNN提供最佳性能和计算效率。该研究强调了根据可用硬件和训练时间与预测时间之间的期望权衡等因素为特定应用选择适当技术的重要性。总体而言，迁移学习技术在图像分类任务中是有效的，特别是当标记数据有限时。
Transfer learning is a popular deep learning technique that involves fine-tuning a pre-trained CNN model on a new dataset to improve accuracy and speed. This article examines the effectiveness of transfer learning techniques in image classification tasks using CNNs. The paper reviews recent studies on transfer learning techniques, including their use in medical image analysis applications such as COVID-19 detection and Alzheimer’s disease classification. The study discusses the ImageNet dataset as a benchmark for pretraining CNN models and proposes an optimized CNN model that uses various optimization techniques to improve performance and efficiency. The article also includes a comparison table of various image classification techniques, including CNN, RNN, SVM, RF, and optimized CNN, with the optimized CNN offering the best performance and computational efficiency. The study emphasizes the importance of selecting the appropriate technique for specific applications based on factors such as available hardware and desired tradeoff between training time and prediction time. Overall, transfer learning techniques are shown to be effective in image classification tasks, particularly when labeled data is limited.