水生生物学报2025，Vol.49Issue(8)：1-12,12.DOI:10.3724/1000-3207.2025.2024.0484
基于麻雀搜索算法优化BP神经网络的叶绿素a浓度反演:以小江为例
OPTIMIZATION OF BP NEURAL NETWORK FOR CHLOROPHYLL-A CONCENTRATION INVERSION BASED ON SPARROW SEARCH ALGORITHM:A CASE STUDY OF XIAOJIANG
摘要
Abstract
Chlorophyll-a concentration is a crucial parameter characterizing water ecological environment quality.To address the issues of traditional Back Propagation(BP)neural networks,which are highly sensitive to initial values and tendency to local optima in chlorophyll-a inversion,this study proposes an SSA-BP inversion model optimized by using the Sparrow Search Algorithm(SSA).A novel inversion model was constructed by integrating remote sensing data from the DJI RTK300 UAV equipped with an AFX-10 hyperspectral camera and synchronous ground sampling data from the Xiaojiang backwater area.The results demonstrate that:(1)The application of Savitzky-Golay(SG)smoothing significantly improved spectral data quality,increasing the determination coefficient(R2)of the SSA-BP model to 0.98;(2)Compared with traditional BP neural networks,the SSA-BP model showed comprehensive improve-ment in inversion accuracy,with the Quma water area exhibiting a 59.14%reduction in Mean Absolute Error(MAE),60.78%decrease in Root Mean Square Error of Prediction(RMSEP),and 57.32%increase in Relative Percent Diffe-rence(RPD);(3)The SSA-BP model overcame the performance degradation of traditional BP models in low-concen-tration regions(where R2 decreased from 0.94 to 0.76),maintaining stable high precision across different chlorophyll-a concentration gradients,with the highest R2 reaching 0.98.This research confirms that the SSA-BP model significantly enhances the accuracy and adaptability of UAV hyperspectral remote sensing in chlorophyll-a inversion,providing a reliable technical approach for ecological environment monitoring in inland water bodies.关键词
麻雀搜索算法/BP神经网络/回水区/叶绿素a浓度/高光谱Key words
Sparrow search algorithm/BP neural network/Backwater area/Chlorophyll a concentration/Hyperspectral分类
农业科技引用本文复制引用
潘炜,李渊,米武娟,段思宇,张羽珩,黄宇波,田楚铭,景晓萱,朱宇轩,毕永红..基于麻雀搜索算法优化BP神经网络的叶绿素a浓度反演:以小江为例[J].水生生物学报,2025,49(8):1-12,12.基金项目
中国长江三峡集团有限公司项目(0711635/0711636)资助[Supported by the China Three Gorges Corporation(0711635/0711636)] （0711635/0711636）
