Different varieties of eggs contain different nutrients and ingredients as a nutritious agricultural product. The phenomenon of inferior quality and adulteration poses a serious threat to food safety, which makes an urgent need to solve the problem of egg variety detection. Four egg varieties as research objects were divided into the training and test sets according to 2∶1 with 160 and 80 eggs respectively. A hyperspectral imaging system was utilized to capture the egg spectral image in the 935.61～1 720.23 nm range. Region of Interest (ROI) with a center size of 30×30 pixels of egg sample was selected after black and white correction, and the average reflectivity of each pixel in the region was extracted as the original spectral data of the sample. The average spectral information in the 949.43～1 709.49 nm range was intercepted for the subsequent study to reduce the influence of random noise at both ends. Savitzky-Golay (SG) smoothing algorithm and multiple scattering correction (MSC) were used to pretreat the effective bands after denoising. The feature wavelengths of the preprocessed spectral data were extracted using a successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS) single screening, and combinations of CARS-SPA and CARS+SPA, respectively. Support vector machine (SVM), particle swarm optimization (PSO) optimized SVM model (PSO-SVM), and extreme learning machine (ELM) model were established based on full bands (FB) and feature band, which were compared to find the best variety classification model. The experimental results showed that the SG-SPA-ELM model has the best identification effect with the best classification accuracy of 85.00%. Hyperspectral imaging technology combined with ELM can effectively realize non-destructive, efficient, and accurate identification of egg varieties and provide references for egg adulteration detection and identification of other agricultural products.