.png){kind=icon}
.png){kind=icon}
Return
Optimizing scene classification: A robust approach with transfer learning and automated machine learning integration
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jer.2025.05.010.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Scene classification is a fundamental challenge in computer vision. Recognizing the complexity of this task is the aim of our study that addresses the need for accurate and robust scene classification by leveraging the capabilities of two widely recognized databases. The motivation behind this research lies in enhancing the accuracy and efficiency of scene classification systems. Therefore, our primary goal is to explore and implement a comprehensive methodology that combines transfer learning and automated machine learning techniques to achieve superior classification results. Our approach commences with a meticulous data loading process, followed by preprocessing steps to ensure the optimal representation of information. We have conducted class distribution analysis to understand the dataset's nuances. Subsequently, we have employed two key models: MobileNetV2 for transfer learning and a custom convolutional neural network (CNN) model featuring batch normalization. This diverse methodology aims to capture intricate patterns within the data. An innovative step of our approach involves employing Tree-based Pipeline Optimization Tool (TPOT), an automated machine learning tool, for model selection and hyperparameter tuning. The results underscore the effectiveness of our methodology, achieving impressive classification accuracy across diverse scenes. This research contributes valuable insights into the integration of transfer learning and automated machine learning for robust and accurate scene recognition, offering a comprehensive approach to address the complexities of scene classification.
Keywords:
CNN
Scene Recognition
TPOT
MobileNetV2
Transfer Learning
Machine Learning
Journal
IF:
2.2
Papers:
371
Citations:
1.7K