Selection method of interval spectrum feature wavelength variables based on improved genetic algorithm
1. School of Information Science and Engineering, Changzhou University, Changzhou, Jiangsu 213164, China; 2. School of Information and Computer, Anhui Agricultural University, Hefei, Anhui 230036, China
Abstract： To improve the robustness and prediction accuracy of the soil nutrient nearinfrared spectroscopy prediction model, a near infrared interval spectrum selection method was proposed based on the improved genetic algorithm. According to the positive and negative change times in the NIRS full spectrum wavelength variables purity gradients of the soil available phosphorus, the full spectrum was divided into multiple wavelength intervals. Using the variable projection importance coefficients (VVIP) from partial least squares regression model (PLSR) output greater than one as extraction criteria, the wavelength intervals with stronger interpretability for predicting soil nutrient target amount were extracted, and the wavelength intervals were combined into an interval spectrum. PLSR was modeled with the interval spectrum feature wavelength variable (FWV), and an improved genetic algorithm was used to select the optimal FWV corresponding to PLSR root mean square error minimum. The experimental results show that the proposed method for selecting optimal FWV can improve the robustness and prediction accuracy of regression model with simplified model structure. The improved real coded differential mutation operator can improve the genetic algorithm to expand the search space of global optimal solution and increase the convergence rate.
