A Hybrid Capsule Network for Hyperspectral Image Classification
Massoud Khodadadzadeh, Xuemei Ding, Priyanka Chaurasia, Damien Coyle
Abstract
Limited training data, high dimensionality, image complexity, and similarity between classes are the main challenges confronting Hyperspectral Image (HSI) classification which may result in suboptimal classification performance. To address such issues, here we introduce the Capsule Network (CapsNet) approach. CapsNet preserves the hierarchy between different parts of the entity in an image by replacing scalar representations with vectors. Motivated by CapsNet, this paper presents a novel end-to-end deep learning (DL) architecture, the Hybrid Capsule Network (HCapsNet), for HSI classification. HCapsNet employs 2D and 3D Convolutional Neural Networks (CNNs) to extract higher-level spatial and spectral features. In order to establish a route between capsules in the lower layers to the most-related capsule in the higher layer, dynamic routing (DR) is used to identify several overlapped objects during training sessions. Hyperparameter optimization is performed using nested cross-validation (Nested-CV) to ensure through generalisiation evaluation. The proposed HCapsNet significantly outperformed the state-of-the-art methods in terms of overall classification accuracy on three widely used hyperspectral datasets, Indian Pines dataset achieving (>3%, p < 110<sup>-8</sup>), the University of Pavia dataset (>4%, p< 110<sup>-6</sup>), the Salinas Valley dataset (>3%, p < 110<sup>-11</sup>) when using only 1% of the data for training. The performance of all CNN-based approaches degraded significantly with smaller training sample sizes. HCapsNet, therefore, offers significant potential in situations with low sample sizes outperforming state-of-the-art methods.