Pol-SAR Image Classification using Edge Semantic Information in Machine Learning

Deep neural networks have achieved significant success in classifying polarimetric synthetic aperture radar (PolSAR) images. However, these images are often affected by speckle noise, leading to reduced spatial consistency and classification accuracy. While semantic segmentation methods can provide detailed classification maps, they typically require large, labeled datasets, which are both costly and time-consuming to obtain. This paper introduces a multifarious fully convolutional network (MFCN) for PolSAR image classification that addresses these challenges by using multi-modal pixel characteristics as inputs. Elastic net classifiers are employed for efficient classification, while principal component analysis (PCA) is used to enhance feature selection and improve robustness. Additionally, edge semantic information is integrated to refine boundary detection, minimize over-smoothing, and increase spatial precision. Extensive experiments on five real-world PolSAR datasets show that the proposed method achieves up to a 5% improvement in classification accuracy compared to existing methods, demonstrating its effectiveness in handling noisy, complex data while maintaining computational efficiency.