Modeling research on wheat protein content measurement using near-infrared reflectance spectroscopy and optimized radial basis function neural network

In this study, near-infrared reflectance spectroscopy and radial basis function (RBF) neural network algorithm were used to measure the protein content of wheat owing to their nondestructiveness and quick speed as well as better performance compared to the traditional measuring method (semimicro-Kjeldahl) in actual practice. To simplify the complex structure of the RBF network caused by the excessive wave points of samples obtained by near-infrared reflectance spectroscopy, we proposed the particle swarm optimization (PSO) algorithm to optimize the cluster center in the hidden layers of the RBF neural network. In addition, a series of improvements for the PSO algorithm was also made to deal with its drawbacks in premature convergence and mechanical inertia weight setting. The experimental analysis demonstrated that the improved PSO algorithm greatly reduced the complexity of the network structure and improved the training speed of the RBF network. Meanwhile, the research result also proved the high performance of the model with its root-mean-square error of prediction (RMSEP) and prediction correlation coefficient (R) at 0.26576 and 0.975, respectively, thereby fulfilling the modern agricultural testing requirements featuring nondestructiveness, real-timing, and abundance in the number of samples.