FusionNet-RS: A Deep Feature Fusion Model for Remote Sensing Image Classification on PatternNet

Remote sensing image classification plays a crucial role in various applications, including land cover mapping and urban planning. Traditional classification methods struggle with complex spatial patterns, necessitating advanced deep learning techniques for improved accuracy and efficiency. Convolutional Neural Networks (CNNs) have demonstrated remarkable success in remote sensing, offering automated feature extraction and superior classification performance. However, selecting an optimal CNN architecture remains challenging due to trade-offs between computational efficiency and accuracy. This study proposes a deep learning model integrating MobileNet and DenseNet121 to enhance classification performance on the PatternNet dataset. MobileNet's lightweight structure ensures computational efficiency, while DenseNet121's deep feature extraction improves classification accuracy. A five-fold cross-validation approach was employed to evaluate the model’s robustness. Experimental results indicate that the feature fusion model achieves a mean accuracy of 95.98%, outperforming individual architectures in both accuracy and generalization.