Abstract
Hyperspectral images (HSIs) are characterized by high spatial resolution and are rich in spectral information. In the process of HSI classification, the extraction of spectral-spatial features directly influences the classification results. In recent years, the hyperspectral classification method based on convolutional neural networks has demonstrated excellent performance. However, as the network structure deepens, degradation occurs, and the features learned from the fixed-scale convolutional kernels are usually specific, which is not conducive to feature learning and thus impairs the classification accuracy. To solve the problem of difficult extraction of features and underutilization of information from HSI data, a densely connected multiscale attention network based on 3-D convolution is proposed for HSI classification. First, to reduce the spectral redundancy of the HSIs, the principal component analysis algorithm is performed on the raw HSI data; then, several multiscale blocks comprised of parallel factorized spatial-spectral convolution modules of different sizes are adopted to extract the enriched spectral-spatial features from HSIs; furthermore, dense connections are introduced to further fuse features obtained from blocks of different depths, thereby enhancing feature reuse and propagation and helping to alleviate the problem of vanishing gradients. Besides, the channel-spectral-spatial attention block is put forward to spontaneously reweight the fused features to emphasize the features that are more relevant to the classification results while weakening the less relevant ones. The experimental results show that the proposed method is effective in extracting discriminative features of the target and outperforms the other state-of-the-art methods.
Highlights
H YPERSPECTRAL remote sensing has become a research hotspot in the field of remote sensing, which has been widely applied in many aspects of earth science, such as agriculture, geology, environment monitoring, and so on [1]–[3]
Both models can extract deep features hierarchically, they suffer from two main deficiencies: First, the input 3-D Hyperspectral images (HSIs) cubes data must be flattened to one-dimensional vector to meet the input requirements of these models, resulting in the loss of spatial information; second, since stack autoencoders (SAEs) and deep belief networks (DBNs) are deep neural networks based on full connection layer, as the network deepens, the number of model parameters will become extremely large
The fluctuations of the values of these three indicators become smaller, indicating that the model becomes more stable. The reason for this result is that the introduction of attention block can adaptively assign different weights to different channels, different spectral features, and different spatial regions in the process of extracting features of HSIs, selectively strengthening important features that are useful for classification, that is, increasing the weight of features that play a relevant role, which helps to improve the accuracy of classification
Summary
H YPERSPECTRAL remote sensing has become a research hotspot in the field of remote sensing, which has been widely applied in many aspects of earth science, such as agriculture, geology, environment monitoring, and so on [1]–[3]. SAEs and DBNs were initially applied to HSI classification tasks Both models can extract deep features hierarchically, they suffer from two main deficiencies: First, the input 3-D HSI cubes data must be flattened to one-dimensional vector to meet the input requirements of these models, resulting in the loss of spatial information; second, since SAEs and DBNs are deep neural networks based on full connection layer, as the network deepens, the number of model parameters will become extremely large. In this situation, CNNs, which have achieved excellent performance on many different computer vision tasks, naturally attract considerable attention.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have