Abstract:In the context of the widespread availability of image data today, accurate and fast image classification remains a challenging task. A novel classification model called IDCNet is proposed in this research. The model is based on the reuse of residual blocks for feature extraction. A convolutional neural network is utilized as the backbone structure of the model. In the branching part, feature reuse residual blocks (FRB) and depth dilated convolutions (DDC) units are incorporated to enhance the flow of information and enrich the diversity of features. As a result, the classification performance of the model is effectively improved. Experimental results demonstrate that the proposed model outperforms other comparative methods on CIFAR-10, CIFAR-100, and Fashion MNIST datasets. This showcases the potential and value of the model in the field of image classification.