摘要
文中针对BP神经网络收敛速度慢和易陷入局部最优的问题,构建了基于波段比值参数的PSO-BP内陆水体叶绿素a浓度估算模型。选择特征波段比值作为BP神经网络输入变量,借助PSO算法优化了BP网络的初始连接权值和阈值,利用机载的CASI高光谱数据进行了相关试验,分析了PSO-BP模型的估算精度。结果表明:(1)PSO-BP模型测试集的决定系数R2为0.87,均方根误差RMSE为2.27 ug/L;(2)与直接以波段反射率为参数的BP模型相比,该模型的均方根误差降低了1.93 ug/L;与波段比值为输入参数的BP模型相比,该模型的均方根误差降低了0.64 ug/L。
Against at the problem of slow convergence and local optimum of BP neural network model,an estimation model of chlorophyll-a in inland water using PSO-BP model based on the band ratio was constructed. The characteristic band ratio was selected as the input parameter of BP neural network and the initial connection weights and thresholds of BP network were optimized by using PSO algorithm. Estimation accuracy of PSO-BP model was analyzed using airborne CASI hyperspectral data for the correlation testing. Results showed that：（ 1） R2 of PSO-BP model was 0. 87,and RMSE（ Root Mean Squared Error） was 2. 27 ug/L;（ 2） Comparing with the BP model which took the band reflectivity as the input parameter directly,RMSE was decreased by 1. 93 ug/L; comparing with the BP model which selected the band ratio as the input parameter,RMSE of estimation result was decreased by 0. 64 ug/L.
作者
罗义平
孙文彬
Luo Yiping;Sun Wenbin(College of Geoscience and Surveying Engineering, China University of Mining ＆ Technology, Beijing , Beijing 100083, China)
出处
《矿山测量》
2018年第3期114-118,共5页
Mine Surveying
基金
中国矿业大学(北京)"越山奇青年学者"资助计划资助
