A Lightweight Random Feature Framework for Efficient Image Classification

Image classification is a fundamental task in modern machine learning, serving as the backbone for a wide range of computer vision applications. Its impact extends across numerous domains, including healthcare, industrial inspection, and environmental monitoring. Despite the remarkable success of deep learning approaches, particularly convolutional architectures, their substantial computational, memory, and energy requirements pose significant challenges for deployment in resource-constrained settings such as edge devices, embedded systems, and real-time applications. To address these limitations, we propose Roc2Img, a lightweight random feature–based approach that extends Rocket kernels from time series to image classification. Roc2Img consists of training a linear classifier on features extracted via random convolutional kernels. Empirical evidence on benchmark and real-world datasets demonstrates that Roc2Img achieves competitive performance while requiring only a fraction of the computational cost of state-of-the-art methods. Code to reproduce the experiments can be found at: https://github.com/Grandi-Luca/roc2img.