Robust 3D Image Recognition via Fractional-Order Laguerre Moments and Deep Residual Networks

To achieve high-speed and high-accuracy recognition of three-dimensional (3D) images while addressing the limitations of traditional feature-extraction methods in robustness and accuracy caused by the high dimensionality and complex structures of 3D data, a robust 3D image recognition approach based on fractional-order Laguerre moment residual blocks is proposed. Specifically, fractional-order Laguerre moments are designed as a feature enhancement module and integrated into a deep residual network to capture the global shape and structural characteristics of 3D images, thereby improving the model’s capacity to extract geometric details and texture features. By combining the Laguerre moment module with convolutional neural network (CNN) features within the residual framework, the proposed method enables joint modeling of local and global information, significantly enhancing noise robustness and generalization ability. The experimental evaluation on the ModelNet40 and PSB datasets shows that the proposed method consistently outperforms state-of-the-art methods in terms of classification accuracy, noise robustness, and generalization ability, with particularly strong performance under low signal-to-noise ratio conditions. This study thus provides a scalable and effective solution for robust and efficient 3D image recognition.