To effectively solve the problems that most convolutional neural networks cannot be applied to the pixelwise input in remote sensing (RS) classification and cannot adequately represent the spectral sequence information, we propose a new multispectral RS image classification framework called HyFormer based on Transformer. First, a network framework combining a fully connected layer (FC) and convolutional neural network (CNN) is designed, and the 1D pixelwise spectral sequences obtained from the fully connected layers are reshaped into a 3D spectral feature matrix for the input of CNN, which enhances the dimensionality of the features through FC as well as increasing the feature expressiveness, and can solve the problem that 2D CNN cannot achieve pixel-level classification. Secondly, the features of the three levels of CNN are extracted and combined with the linearly transformed spectral information to enhance the information expression capability, and also used as the input of the transformer encoder to improve the features of CNN using the powerful global modelling capability of the Transformer, and finally the skip connection of the adjacent encoders to enhance the fusion between different levels of information. The pixel classification results are obtained by MLP Head. In this paper, we mainly focus on the feature distribution in the eastern part of Changxing County and the central part of Nanxun District, Zhejiang Province, and conduct experiments based on Sentinel-2 multispectral RS images. The experimental results show that the overall accuracy of HyFormer for the study area classification in Changxing County is 95.37% and that of Transformer (ViT) is 94.15%. The experimental results show that the overall accuracy of HyFormer for the study area classification in Nanxun District is 95.4% and that of Transformer (ViT) is 94.69%, and the performance of HyFormer on the Sentinel-2 dataset is better than that of the Transformer.

Convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have gained improved results in remote sensing image data classification. Multispectral image classification can benefit from the rich spectral information extracted by these models for land cover classification. This paper proposes a classification model called a hierarchical convolutional recurrent neural network (HCRNN) to combine the CNN and RNN modules for pixel-level classification of multispectral remote sensing images. In the HCRNN model, the original 13-band information from Sentinel-2 is transformed into a 1D multispectral sequence using a fully connected layer. It is then reshaped into a 3D multispectral feature matrix. The 2D-CNN features are extracted and used as inputs to the corresponding hierarchical RNN. The feature information at each level is adapted to the same convolution size. This network structure fully leverages the advantages of CNNs and RNNs to extract temporal and spatial features from the spectral data, leading to high-precision pixel-level multispectral remote sensing image classification. The experimental results demonstrate that the overall accuracy of the HCRNN model on the Sentinel-2 dataset reaches 97.62%, which improves the performance by 1.78% compared to the RNN model. Furthermore, this study focused on the changes in forest cover in the study area of Laibin City, Guangxi Zhuang Autonomous Region, which was 7997.1016 km², 8990.4149 km², and 8103.0020 km² in 2017, 2019, and 2021, respectively, with an overall trend of a small increase in the area covered.

Infrared small target detection technology plays a crucial role in various fields such as military reconnaissance, power patrol, medical diagnosis, and security. The advancement of deep learning has led to the success of convolutional neural networks in target segmentation. However, due to challenges like small target scales, weak signals, and strong background interference in infrared images, convolutional neural networks often face issues like leakage and misdetection in small target segmentation tasks. To address this, an enhanced U-Net method called MST-UNet is proposed, the method combines multi-scale feature decomposition and fusion and attention mechanisms. The method involves using Haar wavelet transform instead of maximum pooling for downsampling in the encoder to minimize feature loss and enhance feature utilization. Additionally, a multi-scale residual unit is introduced to extract contextual information at different scales, improving sensory field and feature expression. The inclusion of a triple attention mechanism in the encoder structure further enhances multidimensional information utilization and feature recovery by the decoder. Experimental analysis on the NUDT-SIRST dataset demonstrates that the proposed method significantly improves target contour accuracy and segmentation precision, achieving IoU and nIoU values of 80.09% and 80.19%, respectively.

Recursive neural networks and transformers have recently become dominant in hyperspectral (HS) image classification due to their ability to capture long-range dependencies in spectral sequences. Despite the success of these sequential architectures, mainstream deep learning methods primarily handle two-dimensional structured data. However, challenges such as the curse of dimensionality, spectral variability, and confounding factors in hyperspectral remote sensing images limit their effectiveness, especially in remote sensing applications. To address this issue, this paper proposes a novel land cover classification algorithm that integrates random forests with a spectral transformer network structure (RS-Net). Firstly, this paper presents a combination of the Gramian Angular Field (GASF) and Gramian Angular Difference Field (GADF) algorithms, which effectively maps the multidimensional time series constructed for each pixel onto two-dimensional image features, enabling precise extraction and recognition in the backend network algorithms and improving the classification accuracy of land cover types. Secondly, to capture the relationships between features at different scales, this paper proposes a SpectralFormer network architecture using the Context and Structure Encoding (CASE) module to effectively learn dependencies between channels. This architecture enhances important features and suppresses unimportant ones, thereby addressing the semantic gap and improving the recognition capability of land cover features. Finally, the final prediction results are determined by a voting mechanism from the Random Forest algorithm, which synthesizes predictions from multiple decision trees to enhance classification stability and accuracy. To better compare the performance of RS-Net, this paper conducted extensive experiments on three benchmark HS datasets obtained from satellite and airborne imagers, comparing various classic neural network models. Surprisingly, the RS-Net algorithm achieves high performance and efficiency, offering a new and effective tool for land cover classification.

This paper proposes a network structure called CAMP-Net, which considers the problem that traditional deep learning algorithms are unable to manage the pixel information of different bands, resulting in poor differentiation of feature representations of different categories and causing classification overfitting. CAMP-Net is a parallel network that, firstly, enhances the interaction of local information of bands by grouping the spectral nesting of the band information and then proposes a parallel processing model. One branch is responsible for inputting the features, normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) band information generated by grouped nesting into the ViT framework, and enhancing the interaction and information flow between different channels in the feature map by adding the channel attention mechanism to realize the expressive capability of the feature map. The other branch assists the network’s ability to enhance the extraction of different feature channels by designing a multi-layer perceptron network based on the utilization of the feature channels. Finally, the classification results are obtained by fusing the features obtained by the channel attention mechanism with those obtained by the MLP to achieve pixel-level multispectral image classification. In this study, the application of the algorithm was carried out in the feature distribution of South County, Yiyang City, Hunan Province, and the experiments were conducted based on 10 m Sentinel-2 multispectral RS images. The experimental results show that the overall accuracy of the algorithm proposed in this paper is 99.00% and the transformer (ViT) is 95.81%, while the performance of the algorithm in the Sentinel-2 dataset was greatly improved for the transformer. The transformer shows a huge improvement, which provides research value for developing a land cover classification algorithm for remote sensing images.

To address the problem that traditional deep learning algorithms cannot fully utilize the correlation properties between spectral sequence information and the feature differences between different spectra, this paper proposes a parallel network architecture land-use classification based on a combined multi-head attention mechanism and multiscale residual cascade called MARC-Net. This parallel framework is firstly implemented by deeply mining the features generated by grouped spectral embedding for information among spectral features by adding a multi-head attention mechanism, which allows semantic features to have expressions from more subspaces while fully considering all spatial location interrelationships. Secondly, a multiscale residual cascade CNN (convolutional neural network) is designed to fully utilize the fused feature information at different scales, thereby improving the network’s ability to represent different levels of information. Lastly, the features obtained by the multi-head attention mechanism are fused with those obtained by the CNN, and the merged resultant features are downgraded through the fully connected layer to obtain the classification results and achieve pixel-level multispectral image classification. The findings show that the algorithm proposed in this paper has an aggregate precision of 97.22%, compared to that of the Vision Transformer (ViT) with 95.08%; its performance on the Sentinel-2 dataset shows a huge improvement. Moreover, this article mainly focuses on the change rate of forest land in the study area. The Forest land area was 125.1143 km${}^2$ in 2017, 105.6089 km${}^2$ in 2019, and 76.3699 km${}^2$ in 2021, with an increase of $15.59\%$, an decrease of $0.97\%$, and increase of $14.76\%$ in 2017–2019, 2019–2021 and 2017–2021, respectively.