<p>One of the most important processes performed on hyperspectral images is their classification. In recent years, convolutional neural networks (CNNs) have been widely used in hyperspectral image classification, each attempting to address the hyperspectral data's computational and processing challenges. Convolutional neural networks become less efficient at solving complex problems as the number of parameters and layers increases. As a result, a new architecture of convolutional neural networks is introduced in this paper, which improves network performance and significantly reduces computing time. The proposed method for reducing spectral bands employs sparse and low-rank representation feature extraction methods based on spectral and spatial information. This model expresses each pixel as a linear combination of dictionary atoms. In addition, the alternating direction multiplier method was used to solve the optimization problem. In this method, the two-dimensional convolutional neural network consists of convolutional, pooling, and fully connected layers. In addition, batch normalization and random elimination are used to prevent overfitting. This study's experiments were conducted using Indiana Pine, Pavia, and Washington DC Mall data sets. The results show that the proposed method has a high classification success rate and a shorter duration, and is less complex than existing models.</p>