Soil Salinity Inversion Model Based on BPNN Optimization Algorithm for UAV Multispectral Remote Sensing

Rapid and accurate inversion of soil salinity is a key scientific problem that needs to be solved urgently. Due to the accuracy of UAV multispectral remote sensing inversion of salinity based on back propagation neural network (BPNN) is low, in this study, used the UAV multispectral image and field measurements of 60 soil surface salinity as data sources, 16 salinity indexs were constructed using the extracted spectral reflectance, and performed a gray relation analysis to screen salinity index features after applying a film removal to construct the BPNN salinity inversion model. Particle swarm optimization (PSO), thinking evolutionary algorithm (MEA), and genetic algorithm (GA) were applied to optimize the BPNN inverse model, respectively, and the optimization capabilities of the four algorithms were compared and evaluated to optimize the best optimization algorithm. The results showed that the GRA variable screening can effectively remove the redundant information of spectral parameters and reduce the complexity of the salinity inversion model; the PSO, MEA, and GA can effectively improve the robusticity of BPNN inversion model, and GA algorithm has the best optimization effect in terms of inverse model optimization effect, followed by MEA and PSO algorithms; the accuracy of the PSO-BPNN, MEA-BPNN, and GA-BPNN inversion models are better than that of the BPNN model, and GA-BPNN is the best salinity inversion model, which achieves R
²
of 0.6659, RMSE of 0.0751, and RPD of 2.0211. This approach can effectively solve salinity monitoring accuracy issues of UAV multispectral inversion.