Remote Sensing Inversion of Suspended Matter Concentration Using a Neural Network Model Optimized by the Partial Least Squares and Particle Swarm Optimization Algorithms

Abstract

Suspended matter concentration is an important index for the assessment of a water environment and it is also one of the core parameters for remote sensing inversion of water color. Due to the optical complexity of a water body and the interaction between different water quality parameters, the remote sensing inversion accuracy of suspended matter concentration is currently limited. To solve this problem, based on the remote sensing images from Gaofen-2 (GF-2) and the field-measured suspended matter concentration, taking a section of the Haihe River as the study area, this study establishes a remote sensing inversion model. The model combines the partial least squares (PLS) algorithm and the particle swarm optimization (PSO) algorithm to optimize the back-propagation neural network (BPNN) model, i.e., the PLS-PSO-BPNN model. The partial least squares algorithm is involved in screening the input values of the neural network model. The particle swarm optimization algorithm optimizes the weights and thresholds of the neural network model and it thus effectively overcomes the over-fitting of the neural network. The inversion accuracy of the optimized neural network model is compared with that of the partial least squares model and the traditional neural network model by determining the coefficient, the mean absolute error, the root mean square error, the correlation coefficient and the relative root mean square error. The results indicate that the root mean squared error of the PLS-PSO-BPNN inversion model was 3.05 mg/L, which is higher than the accuracy of the statistical regression model. The developed PLS-PSO-BPNN model could be widely applied in other areas to better invert the water quality parameters of surface water.