The vision transformer (ViT) has become a hot topic in image processing due to its global feature extraction capabilities. However, the ViT suffers from over-smoothing in feature extraction and over-fitting in the training procedure, so it is hard to achieve satisfactory performance in hyperspectral image (HSI) classification. To address these issues, we propose a ViT with contrastive learning (CViT). The network architecture includes a patch embedding module, transformer blocks, and a classifier. The training of CViT can be considered as an optimization problem with a supervised contrastive loss, an unsupervised contrastive loss, and an ℓ ₁ -regularizer with respect to linear self-attention weights. Specifically, the supervised contrastive loss is proposed to alleviate the negative effects of HSI features’ spectral variability and spatial diversity by increasing intra-class consistency. On the other hand, the unsupervised contrastive loss is exploited to reduce redundancy by reconstructing global structural information. In particular, regularized linear self-attention weights reduce the over-smoothing issue. Extensive experimental results on three HSI datasets demonstrate that the proposed CViT achieves competitive performance.

A novel spectral–spatial difference convolution network (S $^{2}$ DCN) is proposed for hyperspectral image (HSI) classification, which integrates the difference principle into the deep learning framework. S $^{2}$ DCN employs a learnable gradient encoding pattern to extract important detail features in spectral and spatial domains, alleviating the information loss caused by the oversmoothing effect in deep feature extraction. Specifically, the feature extraction modules in S $^{2}$ DCN are designed, namely spectral difference convolution (SeDC) module and spatial difference convolution (SaDC) module. The SeDC module performs 1-D difference convolution in the spectral domain to capture peak-valley information in sensitive narrow bands, enhance subtle spectral differences, and preserve fine-grained features. The SaDC module employs 2-D difference convolution in the spatial domain, integrating fine-structural features while preserving the deep abstract features extracted by vanilla convolutions. This further empowers the capability of the model to extract discriminative features. A series of experiments are performed on four publicly available HSI datasets to demonstrate the effectiveness of S $^{2}$ DCN method, which is compared with current state-of-the-art models. The experimental results show that the proposed S $^{2}$ DCN outperforms competitors and achieves optimal classification performance.

Pixel‐wise classification of hyperspectral image (HSI) is a hot spot in the field of remote sensing. The classification of HSI requires the model to be more sensitive to dense features, which is quite different from the modelling requirements of traditional image classification tasks. Cycle‐Multilayer Perceptron (MLP) has achieved satisfactory results in dense feature prediction tasks because it is an expert in extracting high‐resolution features. In order to obtain a more stable receptive field and enhance the effect of feature extraction in multiple directions, we propose an MLP‐like model called DriftNet for HSI classification inspired by Cycle‐MLP and deformable convolution. Besides, the proposed DriftNet uses a unique ladder‐like fully connected structure to achieve progressive activation of neurons and facilitates the fusion of spatial and spectral information, thereby obtaining more sensitive location information for better classification results. Experimental results on three public hyperspectral datasets demonstrate the effectiveness and generalisation of DriftNet.

Well log data play a crucial role in lithology identification, given their nature as petrophysical measurements of stratigraphy that reflect lithological successions and depositional processes. Extracting recognizable features from well logs has become a focal point in lithology identification tasks. Recent advancements in machine learning have improved feature extraction by incorporating neighborhood information, particularly through filtering techniques that integrate sequence dependencies from well logs. However, traditional filters are often constrained by predefined theoretical models, posing adaptability challenges across varied reservoir conditions. To address these limitations, we propose a novel Category-wise Sequence Correlation (CSC) filter for lithology identification, which is data-driven and model-free. Notably, the CSC filter captures the sequence structural variations arising from category heterogeneity by utilizing labeling information derived from well logs. Building on this, the Category-wise Sequence Correlation Filter Banks (CSCFB) model is developed. It uses a series of CSC filter banks as one-dimensional convolution templates acting on each logging curve to extract discriminative features associated with lithological categories. Application studies on several oil and gas fields demonstrate that the CSCFB model outperforms competitors in classification performance, proving its effectiveness and utility across diverse reservoir types.