Abstract: Due to the simple network structure of the autoencoder, it is difficult to capture the local features of complex images, which significantly reduces the clustering effect of images. To address this situation, unsupervised image clustering algorithm embedded in hybrid-autoencoder (UCHAE) is proposed. The algorithm extracted image features by convolutional autoencoder and autoencoder respectively for adaptive fusion, converted the fused features into a new feature space by orthogonal transformation matrix, and continuously optimized the clustering results according to the target loss function. The UCHAE algorithm was applied to image datasets, namely MNIST, FashionMNIST, COIL-20 and COIL-100, and the clustering accuracy reached 96.5%, 65.8%, 74.5% and 72.0%, respectively. The experimental results show that the algorithm can effectively improve the clustering effect of images.