A Novel Framework of CNN Integrated with Adaboost for Remote Sensing Scene Classification

Deep learning is a powerful means to recognize remote sensing image scene categories. In this study, a deep convolutional neural network (CNN) based ensemble method is proposed. Firstly, a CNN architecture composed of the feature layer and the classifier layer is designed. Then the classifier layer of CNN is treated as base-learner and integrated with the AdaBoost technique to construct a CNN-AdaBoost ensemble framework. The proposed method is compared with the CNN-SVM and fine-tuned VGG16. The experiment results on UC Merced land-use dataset show that the CNN-AdaBoost achieves an improved overall accuracy by 4.46% against the sole CNN. Also, our method outperforms another two paradigms. Therefore, the proposed CNN based ensemble method is promising for image representations regarding remote sensing image scene classification.