**5. Conclusions**

The paper presents a novel two-streams residual network architecture for the classification of hyperspectral data. Our network improves the spectral and the spatial feature extraction by applying a full pre-activation sRN and saRN separately. These two networks are similar in their structure but use a different type of convolutional layer. The convolutional layer of sRN is based on 1D convolution, which best fits the spectral data structure, while the saRN is based on 2D convolution, which best fits the spatial data structure of HSI.

Our experiments were conducted on three well-known hyperspectral datasets, versus five different methods, as well as various sizes of training samples. One of the main conclusions that arises from our experiments is that our proposed method can provide a higher performance versus state-of-the-art classification methods, even with various training samples proportion from 4% training samples up to 30% training samples. The high accuracy of our proposed method on small training samples, 4%, shows that this method does not overfit. Otherwise, the competitive accuracy of our proposed method with large training samples, 30%, explains that this architecture is not under-fitting either.

**Author Contributions:** W.N.K. proposed the methodology, implemented it, did experiments and analysis, and wrote the original draft manuscript. M.B., F.B., and F.S. supervised the study, directly contributed to the problem formulation, experimental design and technical discussions, reviewed the writing, and gave insightful suggestions for the manuscript. D.E. supervised the study, reviewed the writing, gave insightful suggestions for the manuscript and provided resources. All authors have read and agreed to the published version of the manuscript.

**Funding:** W.N. Khotimah is supported by a scholarship from Indonesian Endowment Fund for Education, Ministry of Finance, Indonesia. This research is also supported by Australia Research Council Grants DP150100294, DP150104251, and Grains Research and Development Corporation Grant UWA2002-003RTX.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
