INVERSION OF TOTAL SUSPENDED MATTER CONCENTRATION IN WULIANGSU LAKE BASED ON SWARM INTELLIGENCE OPTIMIZATION AND BP NEURAL NETWORK

Total suspended matter (TSM) is an important parameter of the water environment. Because of the optical complexity in water body, it is difficult to accurately invert the TSM concentration by current traditional methods. In this paper, using Sentinel-2 remote sensing images as the data source combined with measured data, taking Wuliangsu Lake as the study area, a new intelligent algorithm combining the adaptive ant colony exhaustive optimization algorithm (A-ACEO) feature selection method with genetic algorithm (GA) optimized back propagation neural network (BPNN) model (GA-BP) is proposed for inversion of TSM concentration. The ant colony algorithm (ACO) is improved to select remote sensing feature bands for TSM concentration by introducing relevant optimization strategies. The GA-BP model is built by optimizing BPNN using GA with the selected feature bands as input and comparing with the traditional BPNN model. The results show that using feature bands selected by the presented A-ACEO algorithm as inputs, can effectively reduce complexity and improve inversion performance of the model, under the condition of the same model, which can provide valuable references for monitoring the TSM concentration in Wuliangsu Lake.