Scholarlink

<p>Accurate wetland land-cover classification using remote sensing is essential for wetland conservation and management. This study developed a hybrid neural network model that integrates the Sparrow Search Algorithm (SSA), Genetic Algorithm (GA), and backpropagation (BP) network for land-cover classification. The hybrid SSA–GA algorithm optimized the BP network structure and parameters. Utilizing Sentinel-2 imagery, the optimized SSAGA-BP model classified land covers in the Jiangsu Dongtai Tiaozini wetland with higher accuracy compared to BP, GA-BP, SVM, and RF classifiers.The SSAGA-BP model achieved an Overall Accuracy (OA) and an Average Accuracy (AA) of approximately 97%, and a Kappa coefficient of 0.9859. Compared with traditional BP and GA-BP methods, the Kappa coefficient improved by 0.153 and 0.0652, respectively. Further tests in the Yellow River Delta and Chongming Island wetlands demonstrated the transferability of the model, with OA above 93%.This study illustrates that the SSAGA-BP model provides a practical framework for optimizing BP neural networks and enhancing wetland land-cover classification accuracy. By integrating SSA, GA, and BP only once in the workflow description, we avoid redundancy while highlighting the hybrid approach’s contribution. The validated SSAGA-BP model offers a robust and versatile tool for large-scale wetland monitoring using remote sensing.</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

Land Cover Classification Algorithm Based on SSAGA-BP Neural Network in the Tiaozini Wetland, Jiangsu, China

  • 期刊论文

  • 发表日期：

    2026年6月3日

查看全文
  • Siyao Wu,
  • Xia Lu,
  • Fei Wang,
  • Meng Liu,
  • Ke Nie,
  • Mengqiong Xu

摘要

Accurate wetland land-cover classification using remote sensing is essential for wetland conservation and management. This study developed a hybrid neural network model that integrates the Sparrow Search Algorithm (SSA), Genetic Algorithm (GA), and backpropagation (BP) network for land-cover classification. The hybrid SSA–GA algorithm optimized the BP network structure and parameters. Utilizing Sentinel-2 imagery, the optimized SSAGA-BP model classified land covers in the Jiangsu Dongtai Tiaozini wetland with higher accuracy compared to BP, GA-BP, SVM, and RF classifiers.The SSAGA-BP model achieved an Overall Accuracy (OA) and an Average Accuracy (AA) of approximately 97%, and a Kappa coefficient of 0.9859. Compared with traditional BP and GA-BP methods, the Kappa coefficient improved by 0.153 and 0.0652, respectively. Further tests in the Yellow River Delta and Chongming Island wetlands demonstrated the transferability of the model, with OA above 93%.This study illustrates that the SSAGA-BP model provides a practical framework for optimizing BP neural networks and enhancing wetland land-cover classification accuracy. By integrating SSA, GA, and BP only once in the workflow description, we avoid redundancy while highlighting the hybrid approach’s contribution. The validated SSAGA-BP model offers a robust and versatile tool for large-scale wetland monitoring using remote sensing.