AUTHOR=Gezawa Abubakar Sulaiman , Liu Chibiao , Jia Heming , Nanehkaran Y. A. , Almutairi Mubarak S. , Chiroma Haruna TITLE=An improved fused feature residual network for 3D point cloud data JOURNAL=Frontiers in Computational Neuroscience VOLUME=Volume 17 - 2023 YEAR=2023 URL=https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2023.1204445 DOI=10.3389/fncom.2023.1204445 ISSN=1662-5188 ABSTRACT=Due to the sparseness of point clouds, point cloud-based approaches have grown increasingly prominent. Volumetric grid-based techniques are among the most effective models by fully retaining data granularity and utilizing point interdependence. However, employing lower-order local estimate functions to close 3D shapes including the piece-wise constant function or the distance field function required a grid with a high resolution to capture detailed features that need massive processing resources. This paper proposes an improved fused feature network, a complete framework that solves shape classification and segmentation tasks using a two-branch technique and feature learning. To simplify the network efficiently, we begin by designing a feature encoding network that uses two different building blocks with layers skips containing batch normalization (BN) and rectified linear unit (ReLU) in between. Utilizing layers in the training phase helps speed up learning and reduce the effect of gradients vanishing since there are few layers through which to propagate. Also, we create a detailed grid feature extraction module which comprises various convolutions blocks accompanied by a max-pooling to represent a hierarchical representation and extract features from the input grid. Max-pooling is used in each of the pooling layer resulting in each spatial dimension having a smaller grid and helps to manage overfitting by gradually lowering the spatial dimension representation, the parameters in the network as well as the amount of processing. To overcome the limitations of the grid size, the local region in every grid sampled a constant number of points using a simple K-points nearest neighbor (KNN) search which aids in learning approximations functions in higher order to better characterize the details features. The proposed method outperforms or is comparable to state-of-the-art approaches in point cloud segmentation and classification tasks. In addition, a study of ablation is presented to show the effectiveness of the proposed method.