研究目的
To improve the performance of DCNN in Polarimetric SAR image classification by introducing superpixel segmentation and input pyramid to address the issues of neglecting neighboring pixels' similarity and fixed size input.
研究成果
The proposed hybrid framework combining SLIC superpixel segmentation and DCNN with an input pyramid significantly improves classification accuracy and efficiency for POLSAR images, demonstrating good boundary adherence and detail preservation.
研究不足
The method's performance may vary with different scenes and the quality of superpixel segmentation. The computational efficiency, while improved, may still be a constraint for very large datasets.
1：Experimental Design and Method Selection:
The study combines SLIC superpixel segmentation with DCNN for POLSAR image classification, introducing an input pyramid to include multi-scale information.
2：Sample Selection and Data Sources:
Two scenes of POLSAR images from ALOS-2 PALSAR-2 are used.
3：List of Experimental Equipment and Materials:
Intel Core i3-2120 CPU, Nvidia GeForce GTX 1080 GPU, Tensorflow library.
4：Experimental Procedures and Operational Workflow:
The POLSAR data is converted to pseudo-RGB image for superpixel clustering; pyramid samples are produced to train the DCNN; the learned DCNN predicts the category of each superpixel.
5：Data Analysis Methods:
Confusion matrix is used to evaluate 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
