Recurrent Convolutional Layers for Feature Extraction in Image Classification

We propose a novel architecture for artificial convolutional neural networks utilizing a recurrent convolutional layer as the initial feature extractor for image classification. Recurrence in the first layer allows flexible control over feature extraction depth, improving classification accuracy for compact networks—especially on datasets with complex structures. Experiments on CIFAR-10 and MNIST demonstrate that the proposed approach enhances generalization and enables a controlled balance between accuracy and runtime.