Scholarlink

Accurate crop classification using Synthetic Aperture Radar (SAR) is crucial for agricultural monitoring and yield estimation. Feature extraction techniques enhance classification by capturing key texture and structural information from SAR images. Methods like thresholding, backscatter-based classification, and machine learning with handcrafted features face challenges such as speckle noise, limited discriminative power, and sensitivity to seasonal changes, reducing accuracy. This study evaluates three feature extraction methods—2D Wavelet Transform, Stationary Wavelet Transform (SWT), and Curvelet Transform—for classifying five crops (banana, mango, papaya, coconut, and guava) in Madurai using Sentinel-1 SAR images. After pre-processing and speckle noise reduction, the images were classified using a Random Forest model. The results showed 2D Wavelet Transform had the highest accuracy: Banana (98.58%), Coconut (87.99%), Guava (95.02%), Mango (94.07%), and Papaya (99.28%), with SWT and Curvelet Transform performing slightly lower. This study highlights the importance of selecting an effective feature extraction technique to improve classification accuracy in SAR-based agricultural monitoring.

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

Impact of Texture Feature Extraction Methods on Machine Learning-Based Agricultural Crop Classification from Sentinel-1 SAR Data

  • 章节

  • 发表日期：

    2026年

查看全文
  • M. Anandbabu,
  • T. S. Yogesh,
  • B. Karthikeyan,
  • R. A. Alagu Raja,
  • V. Kumar

摘要

Accurate crop classification using Synthetic Aperture Radar (SAR) is crucial for agricultural monitoring and yield estimation. Feature extraction techniques enhance classification by capturing key texture and structural information from SAR images. Methods like thresholding, backscatter-based classification, and machine learning with handcrafted features face challenges such as speckle noise, limited discriminative power, and sensitivity to seasonal changes, reducing accuracy. This study evaluates three feature extraction methods—2D Wavelet Transform, Stationary Wavelet Transform (SWT), and Curvelet Transform—for classifying five crops (banana, mango, papaya, coconut, and guava) in Madurai using Sentinel-1 SAR images. After pre-processing and speckle noise reduction, the images were classified using a Random Forest model. The results showed 2D Wavelet Transform had the highest accuracy: Banana (98.58%), Coconut (87.99%), Guava (95.02%), Mango (94.07%), and Papaya (99.28%), with SWT and Curvelet Transform performing slightly lower. This study highlights the importance of selecting an effective feature extraction technique to improve classification accuracy in SAR-based agricultural monitoring.