This paper proposes a 3D convolutional neural network (3D-CNN) framework for hyperspectral image (HSI) classification. The method directly processes 3D HSI data without preprocessing or post-processing, effectively extracting deep spectral-spatial features. It requires fewer parameters than other deep learning methods, making it more efficient and less prone to overfitting. The proposed 3D-CNN outperforms existing methods such as stacked autoencoder (SAE), deep belief network (DBN), and 2D-CNN on three real-world HSI datasets. Experimental results show that the 3D-CNN achieves the highest overall accuracy (OA) on all datasets, with the best performance on the Indian Pines scene (99.07% OA), significantly outperforming 2D-CNN (95.97% OA). The 3D-CNN also demonstrates better performance in handling spectral-spatial features, capturing local 3D patterns that enhance classification accuracy. The method is lightweight, easy to train, and effective in exploiting both spectral and spatial information in HSI data. Future work includes exploring more effective 3D-CNN techniques that can utilize unlabeled samples.This paper proposes a 3D convolutional neural network (3D-CNN) framework for hyperspectral image (HSI) classification. The method directly processes 3D HSI data without preprocessing or post-processing, effectively extracting deep spectral-spatial features. It requires fewer parameters than other deep learning methods, making it more efficient and less prone to overfitting. The proposed 3D-CNN outperforms existing methods such as stacked autoencoder (SAE), deep belief network (DBN), and 2D-CNN on three real-world HSI datasets. Experimental results show that the 3D-CNN achieves the highest overall accuracy (OA) on all datasets, with the best performance on the Indian Pines scene (99.07% OA), significantly outperforming 2D-CNN (95.97% OA). The 3D-CNN also demonstrates better performance in handling spectral-spatial features, capturing local 3D patterns that enhance classification accuracy. The method is lightweight, easy to train, and effective in exploiting both spectral and spatial information in HSI data. Future work includes exploring more effective 3D-CNN techniques that can utilize unlabeled samples.