A new deep convolutional neural network for fast hyperspectral image classification

By Mercedes Eugenia Paoletti, Juan Mario Haut, Javier Plaza, Antonio J. Plaza. Isprs Journal of Photogrammetry and Remote Sensing, 2017

563 citations recorded by the source corpus.

Abstract Artificial neural networks (ANNs) have been widely used for the analysis of remotely sensed imagery. In particular, convolutional neural networks (CNNs) are gaining more and more attention in this field. CNNs have proved to be very effective in areas such as image recognition and classification, especially for the classification of large sets composed by two-dimensional images. However, their application to multispectral and hyperspectral images faces some challenges, especially related to the processing of the high-dimensional information contained in multidimensional data cubes. This results in a significant increase in computation time. In this paper, we present a new CNN architecture for the classification of hyperspectral images. The proposed CNN is a 3-D network that uses both spectral and spatial information. It also implements a border mirroring strategy to effectively process border areas in the image, and has been efficiently implemented using graphics processing units (GPUs). Our experimental results indicate that the proposed network performs accurately and efficiently, achieving a reduction of the computation time and increasing the accuracy in the classification of hyperspectral images when compared to other traditional ANN techniques.

Subfields: Atmospheric Science, Media Technology

View Paper: https://www.semanticscholar.org/paper/86fdd7fe5700f00e4bcc70983a97c1b9669c953b

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Atmospheric Science

In Media Technology

Citations in the corpus

Deep Pyramidal Residual Networks for Spectral–Spatial Hyperspectral Image Classification (2019, methodology): “4) Finally, the fourth experiment compares the proposed approach with the DFCNN network using three different spatial sizes, 9×9, 15×15, and 19×19, and we use the training configuration considered in [50].”
Hyperspectral Image Classification—Traditional to Deep Models: A Survey for Future Prospects (2021, methodology): “An efficient deep 3D-CNN framework was proposed in [151] that simultaneously exploits”
Deep Learning Meets Hyperspectral Image Analysis: A Multidisciplinary Review (2019, background): “Concerning computational speed, in [214], Paoletti et al. proposed an implementation of 3-”
Extended Vision Transformer (ExViT) for Land Use and Land Cover Classification: A Multimodal Deep Learning Framework (2023, background): “For example, Paoletti et al. [19] made an earlier attempt of constructing a novel spatial–spectral convolutional neural network (CNN) with efficient implementation for fast classification of HS images.”
A review of neural networks in plant disease detection using hyperspectral data (2018, background): “by a set of blocks that can be applied both across space and across time [92].”
A review of deep learning methods for semantic segmentation of remote sensing imagery (2021, background): “Paoletti et al. (2018) also developed a 3D CNN network that is capable of processing both spectral and spatial features of the hyperspectral image simultaneously.”
Deep Learning for Hyperspectral Image Classification: An Overview (2019, methodology): “Based on this fact, 3-D CNN was used to effectively extract deep spectral–spatial-combined features for accurate HSI classification without relying on any preprocessing or postprocessing techniques [56]–[58], [92]–[95].”
Capsule Networks for Hyperspectral Image Classification (2019, methodology): “In addition, Tables VII and VIII give an experimental comparison among the 3-D CNN [24], DFCNN, [29] and the proposed approach using the IP and UP data sets and considering multiple input spatial sizes.”
Deep Few-Shot Learning for Hyperspectral Image Classification (2019, methodology): “D CNN [42]–[44] have also been used for HSI classiﬁcation and have achieved promising classiﬁcation results.”