研究目的
Investigating the application of 3D CNN in terrain classification of PolSAR data, including designing a new convolutional neural network architecture and using the elements of polarimetric coherency matrix as input data.
研究成果
The proposed 3D CNN architecture demonstrates better performance in PolSAR terrain classification compared to 2D CNN, especially in scenarios with less training data or more classes. The method effectively utilizes multichannel information from the polarimetric coherency matrix, leading to higher classification accuracy.
研究不足
The study does not discuss the computational complexity and training time of the proposed 3D CNN architecture compared to traditional methods. Additionally, the impact of varying the size of the input patches and the number of training samples on classification accuracy is not explored.
1：Experimental Design and Method Selection:
The study involves designing a 3D CNN architecture for PolSAR data classification, utilizing the elements of the polarimetric coherency matrix as input.
2：Sample Selection and Data Sources:
Two real PolSAR datasets from San Francisco and Flevoland areas are used for validation experiments.
3：List of Experimental Equipment and Materials:
NASA/JPL AIRSAR polarimetric SAR data.
4：Experimental Procedures and Operational Workflow:
The process includes training the network with sample blocks and labels, performing 3D convolution, pooling, nonlinear operations, and back-propagation, and then classifying the whole image using the trained model.
5：Data Analysis Methods:
The performance of 3D CNN is compared with 2D CNN and BP neural network 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
