研究目的
To overcome the overfitting problem in SAR target recognition due to limited training data by using Gabor features for data augmentation and designing a deeper DCNNs.
研究成果
The proposed G-DCNNs with data augmentation using Gabor filters significantly improves SAR target recognition accuracy, achieving 96.32% on the MSTAR dataset, outperforming traditional methods.
研究不足
The study does not discuss the computational complexity or the time required for training the G-DCNNs with the augmented dataset.
1：Experimental Design and Method Selection:
The study employs Gabor filters for data augmentation to extract multi-scale and multi-direction features from raw SAR images, followed by the design of a G-DCNNs for target recognition.
2：Sample Selection and Data Sources:
The MSTAR public dataset is used, with SAR images cropped to 64×64 patches.
3：List of Experimental Equipment and Materials:
Not explicitly mentioned.
4：Experimental Procedures and Operational Workflow:
Raw SAR images are transformed into 36 Gabor feature maps per image, which are then used to train the G-DCNNs. The network architecture includes convolutional layers, batch normalization, ReLU activation, max pooling, and a global average pooling layer.
5：Data Analysis Methods:
The performance is evaluated based on recognition accuracy on the MSTAR dataset, comparing with traditional methods.
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
