Despite the proven effectiveness of deep learning technology in pine tree disease segmentation, acquiring a large volume of labeled data remains challenging and inefficient. Few-shot segmentation (FSS) uses a small amount of labeled data to guide the segmentation of unknown categories, further evolving into one-shot segmentation (OSS), which utilizes a single labeled sample to perform segmentation under conditions of extreme data scarcity. However, these methods are mostly applicable to natural images with clear boundaries and have not yet been applied to segmenting pine tree disease in autonomous aerial vehicle (AAV) remote sensing images. For this reason, we have designed the OSS model C2Net for the first time, which includes two main modules: 1) a prior contour awareness module (PCAM) that first generates a query image prior mask with contour response and then uses an iterative feature refinement unit (FRU) to refine features and accurately delineate the segmentation boundaries of pine tree disease and 2) an interclass relationship module (ICRM), which studies the vegetation index features of the support and query images, constructing importance weights that reflect the differences between categories, solving the visual similarity issue. Our experiments on field-collected and publicly available datasets demonstrate that C2Net excels in challenging OSS tasks, showing its ability to generalize across different sensor domains and various disease categories. Especially, on the field acquisition dataset, using just a single labeled pine tree disease image achieves an intersection over union (IoU) of 55.24% and an$F_{1}$of 71.24%.

Hyperspectral unmixing (HU) can obtain subpixel-level ground object information, which is crucial for the fine advancement of imaging spectrum processing technology. Deep learning (DL) has been widely used in HU recently because of its ability to deeply mine complex relevant features in data. Existing DL unmixing methods usually operate only in the original spatial-spectral feature domain. However, due to noise, spectral variation, and other factors, it is difficult to fully mine effective features and easy to interfere with by only relying on the original domain. To get over these obstacles, we propose an innovative stationary wavelet convolutional network (SWC-Net) for HU. Stationary wavelet transform (SWT) is introduced in SWC-Net to extend the original feature domain to feature domains with different frequencies, which promotes the multiview extraction of information. What is more, a new generative self-supervised feature learning strategy based on wavelet perspective (GSFL-W) is proposed for SWC-Net. More robust features can be obtained by GSFL-W by introducing noisy perturbations into high-frequency inputs and forcing the network to generate the original inputs. The proposed SWC-Net surpasses the advanced approaches by sufficient experiments on one simulated and three real hyperspectral datasets. The code is publicly available athttps://github.com/UPCGIT/SWC-Net.

Due to the limitations of imaging environment and hardware conditions, mixed pixels are common in hyperspectral images, which seriously affects the accuracy of land use coverage mapping. Subpixel mapping (SPM) decomposes mixed pixels to obtain the spatial distribution information of local object components inside the pixel, thereby breaking through the limitations of traditional pixel-level classification and achieving more accurate land use interpretation and refined mapping. Recently, deep convolutional neural networks have demonstrated their potential and effectiveness in SPM. However, in the SPM process, the multiscale spatial context information are not fully utilized in the process of using semantic information for network modulation, and the spatial representation at a more abstract level cannot be fully obtained. Therefore, in response to the above problems, this article proposes a multiscale semantic modulation hybrid convolutional network for SPM. The network obtains multiscale semantic information in semantics by constructing a multiscale semantic modulation module (MSSM) to modulate the backbone network and fully mine the spatial context information. Simultaneously, a hybrid convolutional module integrating, 2-D convolutional neural networks, 3D convolutional neural networks, and attention mechanisms is designed. This module captures joint spatial–spectral features while reducing model complexity and learns more abstract spatial representations to enhance the network’s performance in SPM. Experimental results show that this method outperforms the most advanced SPM methods on three public datasets and a produced wetland dataset, and the details of land cover categories are more prominent. The code and data will be released on GitHub upon acceptance:https://github.com/UPCGIT/MSMCNet

Real-time Bidding (RTB) advertisers wish to know in advance the expected cost and yield of ad campaigns to avoid trial-and-error expenses.However, Campaign Performance Forecasting (CPF), a sequence modeling task involving tens of thousands of ad auctions, poses challenges of evolving user interest, auction representation, and long context, making coarse-grained and static-modeling methods sub-optimal.We propose AdVance, a time-aware framework that integrates local auction-level and global campaign-level modeling.User preference and fatigue are disentangled using a timepositioned sequence of clicked items and a concise vector of all displayed items.Cross-attention, conditioned on the fatigue vector, captures the dynamics of user interest toward each candidate ad.Bidders compete with each other, presenting a complete graph similar to the self-attention mechanism.Hence, we employ a Transformer Encoder to compress each auction into embedding by solving auxiliary tasks.These sequential embeddings are then summarized by a conditional state space model (SSM) to comprehend long-range dependencies while maintaining global linear complexity.Considering the irregular time intervals between auctions, we

Vision transformers are widely used in hyperspectral image classification, with their core feature extractor being self-attention. Self-attention has a wider receptive field than convolution. However, existing vision transformers for the classification of hyperspectral images (HSIs) with a large number of bands generally suffer from high computational complexity and a large number of parameter requirements. In this paper, we propose an Ultra-lightweight Feature-compressed Multi-head Self-attention Learning Network (UFMS-LN), which mainly consists of a novel Compressed Feature Multi-Head Self-Attention (CF-MHSA), a Spatial Feature Enhancement- Enhancing Transformation Reduction (SFE-ETR) and a Spatial-spectral Hybridization-Receptive Field Attention Convolutional operation (SH-RFAConv). By effectively compressing feature maps in spatial-spectral dimensions, CF-MHSA achieves the same feature extraction capabilities as state-of-the-art self-attention mechanisms, and its floating-point operations (FLOPs) and parameters are two orders of magnitude lower than state-of-the-art self-attention mechanisms. SH-RFAConv is designed to emphasize local features, which have the ability to extract both spatial-spectral features simultaneously and have a wider receptive field than traditional convolutional operations. Furthermore, SFE-ETR is a preprocessing module for UFMS-LN that combines global spatial feature enhancement methods with Enhancing Transformation Reduction (ETR). Extensive experiments conducted on four benchmark HSI datasets have shown that this method achieves superior results compared to existing state-of-the-art HSI classification networks.

The sparse unmixing (SU) technique is widely used in hyperspectral image (HSI) unmixing because it does not need to estimate the number of pure endmembers but directly obtains the spectra from known spectral libraries to construct the endmember matrix, which avoids the influence of endmember extraction on unmixing. However, some existing SU algorithms still have problems, such as insufficient consideration of abundance details and sensitivity to noise. In order to solve the above issues, this article proposes a graph Laplacian weighted robust SU (RSU) algorithm based on superpixels, which can better reconstruct abundance details and reduce sensitivity to noise. The coarse abundance is calculated based on the superpixel results, and then the global spatial prior weight is calculated. Then, weighted RSU is applied to each superpixel to achieve a combination of local and global cooperation to reduce sensitivity to noise. On this basis, in order to better reconstruct the abundance details, the spatial position information and spectral information between pixels within superpixels are used to construct a weighted map to represent the similarity between pixels. Finally, the alternating direction multiplier method (ADMM) is used to perform structural optimization on the superpixel scale, retaining the structural information of abundance and reducing the amount of calculation. Experiments are conducted on three simulated datasets and three real datasets, and the results show that the proposed algorithm outperforms state-of-the-art SU algorithms.

Due to the low resolution of hyperspectral images, the problem of mixed pixels is common, and hyperspectral unmixing is a crucial technology to solve the problem of mixed pixels. Among them, nonnegative matrix factorization (NMF) is widely used because it can simultaneously perform endmember and abundance estimations. As a variant of NMF, the archetype analysis (AA) is to find the most representative sample in the dataset, which has strong interpretability compared with NMF. However, traditional AA-based unmixing methods consider only one spectral curve to represent one class, ignoring endmember variability. To solve this problem, a manifold regularized sparse AA unmixing method considering endmember variability is proposed. In this paper, various spectra were included for each class to fully account for variability. In addition, considering the sparsity of abundance, L2,1regularization is used to impose sparse constraints on abundance, which ensures the sparseness of abundance. Furthermore, a manifold regularization constraint is introduced to use the underlying manifold structure of the data in unmixing, the construction of which is done by superpixel segmentation. The close relationship between the original image and the abundance is preserved. Experimental results on both synthetic and real hyperspectral datasets illustrate that the proposed method is superior to several multi-endmember extraction algorithms, AA-based algorithms, and advanced sparse NMF-based algorithms.

Autoencoders are widely utilized in hyperspectral unmixing as an unsupervised end-to-end learning model. In particular, convolutional autoencoder networks are popular for processing multidimensional hyperspectral features. Nonetheless, the traditional convolutional Autoencoder network’s receptive field is constrained in the unmixing task, and establishing the connection between the local spatial neighborhood and the local spectrum fails to improve unmixing performance significantly. To address these limitations, a bilateral global attention network based on both spatial and spectral information is proposed. It enables the network to obtain respective feature dependencies in the two dimensions and achieve optimal fusion of both features. The network comprises two information extraction branches. The spatial information extraction branch uses the Swin Transformer block to acquire the global spatial attention of the overall image, while the spectral information extraction branch designates a simplified spectral channel attention mechanism to gain spectral attention weight maps. The network’s efficacy is demonstrated through a comparative study using a synthetic dataset and two real datasets. The code of this work is available at https://github.com/UPCGIT/SSABN.

Superpixel segmentation has emerged as a prominent approach for simultaneous extraction of spatial-spectral features in hyperspectral imagery, exhibiting considerable efficacy in this domain. Although effective in spatial spectrum feature extraction, the existing feature extraction algorithms typically perform superpixel segmentation on a single band, failing to utilize the rich spectral and spatial information available across more bands. Moreover, current superpixel feature extraction methods lack scientific guidance for determining optimal multiscale parameters, which can lead to suboptimal segmentation and increased complexity of hyperspectral analysis. To overcome these limitations, this paper presents a novel band-by-band adaptive multiscale superpixel feature extraction method (BAMS-FE). The method comprises of two key components: a band-by-band superpixel-based feature extraction method and an adaptive optimal superpixel multiscale determination method. Firstly, the band-by-band superpixel-based feature extraction method performs superpixel segmentation for each band of hyperspectral images, thereby extracting joint spatial and spectral features. Secondly, the adaptive optimal superpixel multiscale determination method uses an unsupervised approach to determine the optimal multiscale superpixel segmentation parameters. Finally, the BAMS algorithm is obtained by combining the above two algorithms. The proposed algorithm is evaluated on five different datasets, and the results demonstrate its excellent precision and stability. With the top 99% principal components post PCA transformation or with raw, unprocessed hyperspectral datasets, stable and satisfactory classification performance is achieved by BAMS. Additionally, we compared its performance with several other state-of-the-art algorithms and found that it outperformed them in terms of accuracy. Our code will be publicly available at https://github.com/UPCGIT/BAMS-FE.

Autoencoders (AEs) are commonly utilized for acquiring low-dimensional data representations and performing data reconstruction, which makes them suitable for hyperspectral unmixing. However, AE networks trained pixel by pixel and those employing localized convolutional filters disregard the global material distribution and distant interdependencies, resulting in the loss of necessary spatial feature information essential for the unmixing process. To overcome this limitation, we propose an innovative deep neural network model named U-shaped transformer network using shifted windows (UST-Net). UST-Net prioritizes spatial information in the scene that is more discriminative and significant by using multi-head self-attention blocks based on shifted windows. Unlike patch-based unmixing networks, UST-Net operates on the complete image, eliminating inconsistencies associated with patches. Moreover, the downsampling and upsampling stages are used to extract HSI feature maps at different scales. This process generates a context-rich and spatially accurate abundance map without losing local details. The experimental results of one synthetic dataset and three real datasets demonstrate that UST-Net significantly outperforms both traditional and several other advanced neural network methods. Our code is publicly available at https://github.com/UPCGIT/UST-Net.

Hyperspectral unmixing (HU) is widely used to process mixed pixels as an essential technology. Among them, the nonnegative matrix factorization (NMF)-based approach is one typical of the blind unmixing techniques, which can achieve endmembers and abundances simultaneously. Considering the physical meaning of the extracted endmembers, the archetypal analysis (AA) method constructs a new matrix decomposition structure with stronger interpretability than NMF. However, AA ignores the significant sparse property of abundance in unmixing. Therefore, we propose the L1sparsity-constrained AA algorithm for HU. To solve the new optimization problem, we explore a new optimization method for optimizing abundance. The alternating direction method of multipliers (ADMM) is used to increase the strong convexity and convergence of the problem. Then the fast gradient method (FGM) instead of traditional gradient descent is used to speed up algorithm convergence. The experimental results in both the synthesized and real datasets show that the proposed method outperforms several sparse NMF-based and AA-based methods.

In order to research the multi-year spatial-temporal distribution of Harmful Algal Bloom (HAB) in the South Yellow Sea of China, NDVI method was used to extract HAB from MODIS images after Cloud removal by a comprehensive threshold method explored in this paper. Then, the growth law and spatial-temporal distribution of HAB are analyzed by the method of standard deviation ellipse and superposition analysis. It shows that the periphery of the radial sand ridge area off the coast of Jiangsu Province is the main birthplace of the HAB. Entering the breeding period, the northward drift speed as well as the diffusion speed of HAB accelerates. Until late June, it invades the coast of Shandong Province. In the early stage of HAB growth, preventing the spreading along the northeast-southwest direction will effectively restrain the HAB.

Cloud interference often occurs in Enteromorpha prolifera (EP) extraction from MODIS images, with the purpose of solving this problem, a pseudo-invariant feature-based relative radiometric correction algorithm was proposed in this paper for cloud detection, and named PIF-RAC. The pseudo- invariant feature pixels were carried out to find the linear relationship of reflectance between target image and reference image in this algorithm. Then, the cloud detection threshold of the target image was corrected by the above established linear relationship and manual cloud detection threshold of the reference image. The experimental results show the automatic cloud detection effect of the proposed algorithm is close to that of the artificial threshold algorithm, which enables to effectively eliminate different kind of cloud interference for EP information from MODIS images. The PIF-RAC is an unsupervised algorithm with a high level of automation, which can be applied on EP disasters remote sensing operational monitoring.

Yellow Sea and Bohai Sea are the most frequent sea fog regions in China. The research on sea fog detection methods is of great significance to sea safety and human production activities. In this paper, the MODIS images are used for sea fog detection experiment from 2016 to 2018. The sea fog detection threshold algorithm is established based on MODIS 1, 2, 3, 5, 17, 26 and 32 bands. The temporal and spatial characteristics of sea fog are analyzed in the Yellow Sea and Bohai Sea.

A method to solve multi-target classification problems with unknown clutter parameters is proposed in this paper. The unknown parameter is estimated and synthesized at each observation, and probability of each hypothesis is updated. Subsequently, the optimal waveform for the next illumination is designed based on NP criteria, and the final decision is made based on the sequential probability ratio testing. Simulated results are presented based on our method and show that the optimal waveform-based sequential testing can be decided through reduction of the average illumination number. Furthermore, results indicate a significant improvement over the non-optimal waveforms.

Radar performance improvement through waveform optimization has been an ongoing topic of research recent years. In this paper, we use the optimal waveform design method to deal with the moving target in the clutter and noise. Neyman-Pearson detector criterion is used to maximize the probability of target detection. The optimal waveform is then designed theoretically corresponding to the velocity of target and clutter/noise power spectrum density. Simple CW signals can produce maximum detectability based on different noise PSD situations. Simulated results are presented based on our method and improvement in image is approached. Finally, the conclusions are drawn based on our analysis and simulations.

In this paper, we combine the waveform design method with the synthetic aperture algorithm to suppress clutter and generate clear microwave images of targets. The linear recursive model is introduced into the SAR operation principle and Kalman filter algorithm is used to estimate target and clutter responses in each azimuth direction based on their states before, which both are assumed to be Gaussian distributions. Optimal waveforms based on NP criteria are designed repeatedly and used as the transmitting signals. A clutter suppression filter is then designed and added to suppress the clutter response while maintaining most of the target response. The simulations show that our algorithm can significantly reduce the clutter response while targets are imaged.

An innovative algorithm to reconstruct complex target-induced azimuth envelope in synthetic aperture radar (SAR) system is proposed. Unlike the assumption in conventional SAR imaging algorithms, target's backscattering coefficient can hardly remain constant when synthetic aperture time is long enough. In order to fully understand the target feature, we extract both amplitude and phase information of target-induced azimuth envelope from SAR raw data. In this paper, we formulate range migration curve (RMC) with a parametric model and implement curve fitting to estimate these parameters. The input pixels of curve fitting process is extracted from range compression result of raw data. Applying the idea of random sample consensus (RANSAC), this algorithm classifies pixels according to the target's RMC they belongs to, and extracts target's feature information at the same time. Experimental results are conducted to validate this algorithm.

In this paper, we proposed a clutter suppression algorithm for SAR imaging due to different target power spectrum density (PSD) and clutter PSD in azimuth direction. The optimal waveform of the SAR system is designed according to the prior-knowledge of the target and clutter, an amplitude limiter set in frequency domain is used to suppress the clutter response and 2-dimentional pulse compression is then used to the SAR imaging. Simulated results are presented based on our method which is shown great improvement in target image over clutter image. Then conclusions are drawn based on our analysis and simulations.

This paper presents a novel and compressive SAR raw data simulator based on CUDA via GPU. The stripmap synthetic aperture radar(SAR) is introduced to model antenna illumination behavior. Compared with conventional raw data simulators, we no longer limit our research interests on a single point target's raw data simulation, but expend it to that of complex scene. In order to compensate the greatly increasing operational time with booming computational complexity, we optimize the process in two aspects. In the first, modern GPUs have the potential for highly parallel calculation, and it makes them much more efficient than CPUs in processing large blocks of data. Therefore, we implement the simulator on CUDA. The second method is to take advantage of symmetry in single raw data matrix based on stripmap mode SAR, and reduce 75 percent computational complexity. By these two effective methods, the simulator experiences an attractive operating time and enjoys high efficiency. Some simulation results prove the simulated raw data is acceptable in accuracy.

In this paper, a theory of two-dimensional transmit-receiver design is presented. Only deterministic targets and random clutters and noises are considered in this model, and we reach two-dimensional optimization based on maximization signal-to-interference-and-noise ratio(SINR). New result focuses on extending the previous one-dimensional waveform design to two-dimensional one, in order to satisfy some practical application like synthetic aperture radar(SAR). Both the theoretic derivation and simulation result proves that 2D radar imaging can gain high SINR with proper transmit-receiver design.