Maize is one of the three major food crops in the world, and the use of substandard seeds that do not meet the national standards will seriously affect the yield of maize crops, so how to identify substandard maize seeds quickly, accurately and efficiently is particularly important. The hyperspectral image system to obtain the 900~1 700 nm spectral curves of 900 “Yuan 3” corn seeds, in which the training set and test set ratio was 3∶2, 540 and 360 seeds respectively. The seeds were treated with an electric blast dryer to obtain corn seed samples with different degrees of damage, and the germination test was completed after collecting the spectra to determine the viability of the seeds. In order to improve the signal-to-noise ratio, the spectral bands of maize seeds in the range of 963.27~1 698.75 nm were intercepted as the effective bands. Standard Normal Variation (SNV) and Multiplicative Scatter Correction (MSC), were used to pre-process the raw spectral data. The Successive Projections Algorithm (SPA) and Competitive Adaptive Reweighted Sampling (CARS) were used to extract feature bands from the pre-processed spectral data, with wavelength reflectance as input matrix X and preset sample categories as output matrix Y. The Support Vector Machine (SVM) was used to model and analyze the data, and the results showed that the MSC-CARS-SVM model was the best model, with a model recognition success rate of 98.33% and a Kappa coefficient of 0.985. Genetic Algorithm (GA) was used to optimize the penalty coefficient c and kernel function parameter g in the SVM, and the model accuracy was improved to 100% for the identification of heat-damaged counterfeit and poor-quality maize seeds. This study provides a new idea and method for rapidly identifying the pseudo-inferior quality of maize seeds and seeds of other crops.