Scholarlink

<p>Speckle noise in synthetic aperture radar (SAR) imagery degrades quality and obscures critical features. This paper presents a hybrid despeckling method—Speckle Reduction and Edge Preserving Network (SRE-Net)—using a Combined Rajan–Haar Wavelet Transform (CRHWT). The Rajan Transform’s spatial sensitivity is integrated with the Haar Wavelet’s multiscale capability. A logarithmic transform converts multiplicative noise to additive, followed by CRHWT decomposition, thresholding, adaptive filtering, inverse transform, and exponential mapping. Evaluated on SAR images with simulated speckle noise (5–40%), SRE-Net achieves high Peak Signal-to-Noise Ratio (PSNR up to 36.85&#xa0;dB), Structural Similarity Index (SSIM up to 0.98), Universal Image Quality Index (UIQI ≈ 0.98), and low normalized Mean Squared Error (MSE ≈ 0.0001). Comparative analysis with traditional filters and transform-based methods shows superior edge, line, and texture preservation. Statistical significance is confirmed using one-way ANOVA and Tukey’s post hoc test. SRE-Net’s robustness and generalization make it applicable to SAR-based agricultural monitoring, terrain analysis, segmentation, classification, and change detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SRE-Net: An Enhanced Speckle Reduction and Edge Preserving Network for SAR Images

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Narasimhan Renga Raajan,
  • E. Ponmani

摘要

Speckle noise in synthetic aperture radar (SAR) imagery degrades quality and obscures critical features. This paper presents a hybrid despeckling method—Speckle Reduction and Edge Preserving Network (SRE-Net)—using a Combined Rajan–Haar Wavelet Transform (CRHWT). The Rajan Transform’s spatial sensitivity is integrated with the Haar Wavelet’s multiscale capability. A logarithmic transform converts multiplicative noise to additive, followed by CRHWT decomposition, thresholding, adaptive filtering, inverse transform, and exponential mapping. Evaluated on SAR images with simulated speckle noise (5–40%), SRE-Net achieves high Peak Signal-to-Noise Ratio (PSNR up to 36.85 dB), Structural Similarity Index (SSIM up to 0.98), Universal Image Quality Index (UIQI ≈ 0.98), and low normalized Mean Squared Error (MSE ≈ 0.0001). Comparative analysis with traditional filters and transform-based methods shows superior edge, line, and texture preservation. Statistical significance is confirmed using one-way ANOVA and Tukey’s post hoc test. SRE-Net’s robustness and generalization make it applicable to SAR-based agricultural monitoring, terrain analysis, segmentation, classification, and change detection.