K Sumathi, S Pramod Kumar, Mahadevaswamy H R, B S Ujwala
JNN College of Engineering, Shimoga, Karnataka, India.
JSS Mahavidyapeetha, Visvesvaraya Technological University, Mysore, Karnataka, India.
MethodsX. 2025 Feb 17;14:103226. doi: 10.1016/j.mex.2025.103226. eCollection 2025 Jun.
Scene classification plays a vital role in various computer vision applications, but building deep learning models from scratch is a very time-intensive process. Transfer learning is an excellent classification method using the predefined model. In our proposed work, we introduce a novel method of multimodal feature extraction and a feature selection technique to improve the efficiency of transfer learning in scene classification. We leverage widely used convolutional neural networks (CNN) for feature extraction, followed by relevant feature selection techniques to enhance the performance of the model and increase computational efficiency. In this work, we have executed the proposed method on the Scene dataset of 6 classes and the AID dataset. Experimental results indicate that the MIFS-based approach reduces computational overhead and achieves competitive or superior classification accuracy. The proposed methodology offers a scalable and effective solution for scene classification tasks, with potential applications in real-time recognition and automated systems.
场景分类在各种计算机视觉应用中起着至关重要的作用，但从零开始构建深度学习模型是一个非常耗时的过程。迁移学习是一种使用预定义模型的优秀分类方法。在我们提出的工作中，我们引入了一种新颖的多模态特征提取方法和一种特征选择技术，以提高场景分类中迁移学习的效率。我们利用广泛使用的卷积神经网络（CNN）进行特征提取，然后采用相关的特征选择技术来提高模型的性能并提高计算效率。在这项工作中，我们在6类场景数据集和AID数据集上执行了所提出的方法。实验结果表明，基于MIFS的方法减少了计算开销，并实现了具有竞争力或更高的分类准确率。所提出的方法为场景分类任务提供了一种可扩展且有效的解决方案，在实时识别和自动化系统中具有潜在应用。
