研究目的
Evaluating the performance of convolutional neural network combined with support vector machine for classifying aerial images based on presence of a vehicle.
研究成果
CNN based approaches outperform HOG-SVM, with CNN-SVM providing slightly better accuracy than CNN alone. Combining better features with a better classifier results in better classification. Future research should evaluate object recognition performance and try transfer learning on this dataset.
研究不足
The CNN-SVM approach requires more training time since both CNN and SVM have to be trained. The features learned by CNN have not been trained for SVM's optimization objective.
1：Experimental Design and Method Selection:
The study combines convolutional neural networks (CNN) with support vector machines (SVM) for image classification. The CNN is used for feature extraction, and the SVM is used for classification based on these features.
2：Sample Selection and Data Sources:
The VEDAI dataset, consisting of aerial images with various backgrounds and vehicles, is used. Data augmentation is applied to increase the number of samples.
3：List of Experimental Equipment and Materials:
MATLAB is used for implementing CNN and SVM.
4：Experimental Procedures and Operational Workflow:
The CNN is trained with the training data, and the last fully connected layer is removed to extract features for SVM. Test samples are fed into the CNN to get features, which are then classified by SVM.
5：Data Analysis Methods:
The performance of CNN-SVM is compared with HOG-SVM and CNN alone based on classification accuracy.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
