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.

A Sequence-Variable Attention Temporal Convolutional Network (SVA-TCN) is proposed for lithology classification based on well log data. This study aims to address the issue that native TCN pays insufficient attention to crucial logging variables and sequence structural features in well log tasks. A novel Sequence-Variable Attention Mechanism module is designed to effectively extract features by adding the bidirectional attention mechanism, which comprises the sequence attention mechanism and the variable attention mechanism. The sequence attention mechanism, acting along the depth dimension, focuses on modeling the multiscale sequence features. The variable attention mechanism, working along the logging variable dimension, contributes to learning the importance of different logging variables and exploring the internal correlations among them. To verify the validity of the model, experiments are conducted in the volcanic reservoir of the Yingcheng Formation in the Xujiaweizi Depression of the Songliao Basin. Seven logging variables sensitive to lithological changes are selected, including compressional wave slowness, density, and compensated neutron logging, etc., to construct a lithology identification model using SVA-TCN. Compared with machine learning and deep learning methods, the SVA-TCN demonstrates a remarkable accuracy of 99.00%, surpassing the accuracy of the comparison methods by 0.37–17.69%. The findings of this study indicate that the SVA-TCN model is well-suited for lithology identification in volcanic reservoirs. It provides more reliable prediction results and exhibits good stability and generalization, offering a new avenue to volcanic lithology identification.

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.

Traditional deep learning models for hyperspectral image (HSI) classification are limited by static activation functions and linear weights, hindering the extraction of complex features effectively. The recent Kolmogorov-Arnold (KA) network, by introducing learnable activation functions, offers greater model flexibility, demonstrating potential to overcome the limitations of traditional models. However, the KA network still faces several challenges in practical applications, particularly in handling high-dimensional spatial-spectral data fusion, managing the complexities of B-spline functions, and meeting high computational demands. To overcome these challenges, we propose the Adaptive Taylor KA Network (ATKAN) for HSI classification. ATKAN applies KA operations to channels while sliding across the spatial domain to process high-dimensional spatial-spectral features. It also employs the Taylor series instead of B-spline functions for activation functions, ensuring smooth approximations and enhancing the ability to model non-linear patterns. Additionally, ATKAN incorporates adaptive feature shifts along the spatial axis, channel grouping, and shuffling mechanisms to improve information processing efficiency and reduce computational consumption. By integrating these characteristics, ATKAN forms a shallow hybrid model capable of effectively handling complex linear and nonlinear spatialspectral features. Experimental results on six HSI datasets show that ATKAN significantly enhances classification accuracy and efficiency compared to existing methods