As one of the primary steps in NIR spectral analysis, effective feature band selection can improve modelling efficiency and model performance. Traditional Characteristic band selection algorithms suffer from long run times and redundant feature selection, making achieving the desired results in practical engineering applications difficult. The Harris Hawk Optimisation (HHO) algorithm has the advantages of simple principles and few parameters, but it also has the shortcomings of low convergence accuracy and easy to fall into local optimum. In this paper, we propose an NIR spectral feature band selection model based on the Improved Harris Hawk Optimisation (IHHO) algorithm based on the HHO algorithm. For the HHO algorithm can only be used to solve optimization problems in continuous space, a discretization strategy is used to modify the HHO algorithm so that it can solve the discrete form of the characteristic waveform selection problem. Considering the poor quality of the initial population of the HHO algorithm, the quality of the initial population is improved using chaotic mapping and opposition-based learning to enhance the global exploration capability of the algorithm; Due to the low convergence accuracy of the HHO algorithm in local search, a new prey energy decay model and jumping strategy are proposed further to enhance the algorithms search capability in local search. The HHO algorithm is perturbed by borrowing the variational approach of genetic algorithm. Support vector machine (SVM) identification models and partial least squares regression (PLSR) models were developed using competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), particle swarm optimization (PSO) algorithms, genetic algorithms (GA), HHO algorithms compared to IHHO algorithms, and four qualitative analysis NIR spectral datasets and two quantitative analysis NIR spectral datasets, respectively. In the qualitative analysis experiments, the average accuracy obtained by the IHHO algorithm improved by 0.83%, 9.55%, 17.65%, and 0%, respectively, concerning the full band, and the average number of characteristic bands was only 9.97%, 2.59%, 1.36%, and 0.59% of the full band. In the quantitative analysis experiments, the average coefficient of determination obtained by the IHHO algorithm was 10.57%, 1.47%, 4.41%, 3.66% and 3.06% higher than the full band, and the average root mean square error was 0.162, 1.266 3, 1.868, 1.869 4 and 0.408 4 lower than the full band, and the average number of characteristic bands was only 9.24%, 10.53% and 0% of the full band. The average number of characteristic bands was only 9.24%, 10.53%, 6.54%, 6.91% and 7.14% of the full band. The experimental results show that the IHHO algorithm can remove redundancy in the selection of feature bands and target the most important ones, and its performance is better than several other selection algorithms. Therefore, the IHHO algorithm has good application prospects.