Year: 2023
Type: conference-paper
Abstract: The significance of image classification, a fundamental task in computer vision, lies in its potential to revolutionize various real-world applications, including object recognition, autonomous vehicles, and medical image analysis. This paper introduces and employs a series of enhancement strategies, particularly in the context of deep learning, to improve the accuracy and generalization capabilities of Convolutional Neural Networks (CNNs) for image classification. The research objective is to enhance the accuracy and generalization capabilities of the model by applying improved CNNs and conducting hyperparameter tuning. To achieve this objective, this paper utilizes transfer learning, L2 regularization, early stopping strategies, and fine-tuned hyperparameters. Specifically, the initial model undergoes pre-training, and a comprehensive hyperparameter tuning process is undertaken. Consequently, the results of this study, conducted on a comprehensive dataset, demonstrate a significant improvement in image classification accuracy, thus highlighting the effectiveness of performance optimization within the context of CNNs. The precision of the model, particularly in distinguishing challenging categories such as streets and buildings, is notably enhanced, further emphasizing the importance of hyperparameter tuning. The experimental outcomes not only confirm the pivotal role of hyperparameter optimization in achieving superior model performance but also underscore the significance of deep learning within this context. In a broader context, this research holds substantial real-world implications, propelling the fields of computer vision and deep learning forward. The optimized model's potential to excel in image classification tasks across diverse categories has far-reaching applications, paving the way for further innovations in the realm of artificial intelligence.
Author: Sifan Yu
Institution: Xidian University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4396218503
