An Efficient and Lightweight Convolutional Neural Network for Remote Sensing Image Scene Classification.

Area of Science:

• Computer Vision
• Machine Learning
• Remote Sensing

Background:

• Remote sensing image scene classification is crucial for image interpretation.
• Current convolutional neural network (CNN) methods face challenges with large parameter counts and computational costs.
• Lightweight CNNs offer efficiency but often sacrifice classification performance.

Purpose of the Study:

• To develop a more efficient and lightweight CNN for improved remote sensing image classification accuracy.
• To address the limitations of existing methods, particularly with small training datasets.
• To enhance feature extraction and fusion techniques for better scene classification.

Main Methods:

• Proposed a bilinear convolutional neural network (CNN) model inspired by fine-grained visual recognition.
• Utilized MobileNetv2 as a lightweight CNN backbone for initial deep feature extraction.
• Implemented a Hadamard product operation for enhanced bilinear feature generation, followed by pooling and normalization for classification.

Main Results:

• The proposed bilinear CNN method achieved higher accuracy on UC Merced, AID, and NWPU-RESISC45 datasets compared to state-of-the-art methods.
• Demonstrated significantly fewer parameters and lower computational costs than existing approaches.
• Showcased improved performance and accuracy through feature fusion with bilinear pooling.

Conclusions:

• The developed lightweight bilinear CNN offers a superior balance between efficiency and accuracy for remote sensing scene classification.
• The method effectively enhances classification performance, especially beneficial for tasks with limited training data.
• This approach holds potential for broad application across various remote sensing image classification tasks.