The quality of wheat varieties is of paramount importance in both agricultural and food industries, as it directly influences product prices within the food sector. The research aims to explore the utilization of deep convolutional neural networks for classifying twelve Indian wheat varieties, using the spatial and spectral features of hyperspectral images of densely packed bulk wheat seeds. The findings demonstrate that incorporating a bilinear interpolation layer before the base architecture of the deep convolutional neural networks (DCNNs), conventionally intended for RGB images, can effectively be adapted to harness hyperspectral images’ spatial and spectral data. The near-infrared hyperspectral imaging system was utilized to capture images of densely packed bulk wheat seeds with a spectral range of 900–1700 nm. A data augmentation technique was employed to increase the dataset size, resulting in 10,800 images (900 images per variety \(\times \) 12 varieties) for classification. The images were pretreated using two preprocessing techniques: Standard Normal Variate (SNV) and Multiplicative Scatter Correction (MSC). The deep convolutional neural networks, including GoogLeNet, ResNet34, and DenseNet121, were used to classify the wheat varieties. Notably, GoogLeNet combined with MSC outperformed other classifiers and achieved an impressive test accuracy of 95.12%. The study’s conclusions strongly indicate that integrating DCNNs with hyperspectral imaging technology presents substantial potential for the non-destructive identification of wheat varieties, instilling confidence in its future applications.