Spatial Attention Guided Residual Attention Network for Hyperspectral Image Classification

Abstract

Hyperspectral image (HSI) classification has become a research hotspot. Recently, deep learning-based methods have achieved preferable performances by which the deep spectral-spatial features can be extracted from HSI cubes. However, in complex scenes, due to the diversity of the types of landcover and the bands in high dimensional, these methods are often hampered by the irrelevant spatial areas and the redundant bands, which results in the indistinguishable features and the restricted performance. In this article, a spatial attention guided residual attention network (SpaAG-RAN) is proposed for HSI classification, which contains a spatial attention module (SpaAM), a spectral attention module (SpeAM), and a spectral-spatial feature extraction module (SSFEM). Based on the spectral similarity, the SpaAM is capable of capturing the relevant spatial areas composed of the pixels of the same category as the center pixel from HSI cube with a novel inverted-shifted-scaled sigmoid activation function. The SpeAM aims to select the bands which are beneficial to the spectral features representation. The SSFEM is exploited to extract the discriminating spectral-spatial features. To facilitate the processes of bands selection and features extraction, two well-designed spatial attention masks generated by the SpaAM are employed to guide the works of the SpeAM and the SSFEM, respectively. Moreover, a spatial consistency loss function is installed to maintain the consistency between the two spatial attention masks so that the network enables the distinction of the relevant features exactly. Experimental results on three HSI data sets show that the proposed SpaAG-RAN model can extract the discriminating spectral-spatial features and outperforms the state-of-the-arts. INDEX TERMSHyperspectral image classification, attention mechanism, deep learning, spatial attention, residual network. NINGYANG LI received the B.S. degree in remote sensing science and technology from Henan Polytechnic University, Jiaozuo, China, in 2019. He is currently pursuing the M.S. degree in software engineering with the

Spectral-spatial inverted residual attention network for hyperspectral image classification with pre-activation

“…This gap indicates that the proposed 3D IRAN, which integrates the inverted residual network and the linear bottleneck blocks, can handle spatial and spectral-spatial features more efficiently. The OA of Scheme 5 is higher than that of Scheme 4, which proves that the spatial attention of PAA module is better than S 3 AM [47]. It reveals that the strategy of designing two SG activation functions before adaptive aggregation is beneficial to follow the feature of each similarity measurement, which is ignored in the S 3 AM.…”

“…In order to demonstrate the classification performance of the proposed network, some other popular attention-based networks, including 3D ResNet [47], patch-wise spectral former (SFP) [40], compact band weighting network (CBW) [42], residual spectral-spatial attention network (RSSAN) [38] and S3AM-Net [43], are conducted for comparison. Each method is trained, validated and tested on the same data sets.…”

S3AM: A Spectral-Similarity-Based Spatial Attention Module for Hyperspectral Image Classification

“…On the basis of a global average pooling layer and a symmetrical multi-layer perceptron, the SE module can collect the global expressive information and map them to the channel attention. The SE module has made appreciable progress on reweighting the spectral bands adaptively [43], [44]. Besides, it was extended to make full use of spatial contextual information [45], [46].…”

“…Considering the different spectral characteristics between the relevant pixels and the interfering pixels, a spectral-similarity-based spatial attention module (S 3 AM) is proposed in this article. Due to the fact that the spectral similarity depended on a single metric tends to possess weak representation [44], the S 3 AM adopts the Euclidean and cosine distances to obtain the spectral similarity. Nevertheless, the immediate similarities of the two original measures are often inexact due to the notorious spectral variability.…”

Multibranch 3D-Dense Attention Network for Hyperspectral Image Classification

“…For example, support vector machine (SVM) [15]- [17], K-nearest neighbor (KNN) [18], Bayesian models [19], and linear discriminative analysis [20]. With the rapid development of deep learning in the field of vision, Auto Encoder [21], Recurrent neural networks (RNNs) [22]- [24], convolutional neural networks (CNNs) [25]- [29], and Graph Neural Network [30]- [33] have also gained the hearts of researchers in the field of HSI classification. To obtain spectral contextual information, Mou et al [22] proposed using RNN models for HSI classification by treating HSI data as sequence data.…”