研究目的
The major objective is to demonstrate the superiority of WSCN as applied to SAR ATR through extensive experiments. It uses the roto-translation scattering convolution network to extract the target scattering energy characteristics of the SAR image, and then utilizes the extracted features to train Gaussian kernel support vector machine (SVM) for classification.
研究成果
The proposed SAR automatic target classification method based on a wavelet-scattering convolution network achieves high classification accuracy under standard operating conditions and shows robustness to configuration variants. However, its performance degrades under significant changes in depression angle. The method reduces the dependency on the number of training samples and the probability of overfitting, making it preferable for SAR ATR tasks.
研究不足
The proposed method is sensitive to significant changes in depression angle, which affects the classification accuracy. Additionally, the method's performance under extended operating conditions (EOC) is lower compared to standard operating conditions (SOC).
1：Experimental Design and Method Selection:
The study employs a wavelet-scattering convolution network (WSCN) with complex wavelet filters over spatial and angular variables for feature extraction, followed by conventional dimension reduction and a support vector machine (SVM) classifier for classification.
2：Sample Selection and Data Sources:
The MSTAR benchmark dataset, containing SAR images of 10 classes of ground military targets, is used for testing under both standard operating conditions (SOC) and extended operating conditions (EOC).
3：List of Experimental Equipment and Materials:
The study utilizes SAR images collected by Sandia National Laboratory (SNL) SAR sensors under the MSTAR project.
4：Experimental Procedures and Operational Workflow:
The proposed method involves feature extraction using WSCN, dimension reduction, and classification using SVM. The performance is evaluated based on classification accuracy under SOC and EOC.
5：Data Analysis Methods:
The classification accuracy is assessed using Percent Correctly Classified (Pcc) and confusion matrices.
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
