The local learning algorithm is introduced into the optimal wavelength selection of near infrared ray hyperspectral imaging of maize seeds. These obtained wavelengths are used to develop a discrimination model coupled with partial least squares discriminant analysis to implement the rapid discrimination of maize seeds using less wavelengths. 256 near infrared ray hyperspectral images between 874~1734 nm wavelengths are acquired using a hyperspectral imaging system for 720 maize seed samples including six varieties. Local learning algorithm is proposed to calculate the weight values of wavelengths, and the optimal wavelengths are selected according to the weight values. The experimental results show that local learning algorithm can effectively select the optimal wavelengths. Using 13 optimal wavelengths, six groups of maize seeds achieve an average purity of 95.97%, which can provide a suitable technical way for the rapid discrimination of maize seeds.