Remote sensing change detection (CD) is often affected by noisy shallow features and insufficient bi-temporal semantic alignment, resulting in low predictive confidence and blurred boundaries. This letter proposes a confidence-enhanced feature refinement network (CE-FRN) to explicitly improve detection reliability during feature decoding. A feature refinement by graph-inspired inference (FRGI) module abstracts deep difference-enhanced features into compact semantic prototypes and propagates reliable semantics to shallow pixels through node-pixel attention, effectively suppressing noise. In addition, a layer feature fusion block (LFFB) integrates bi-temporal semantic and difference cues under confidence-aware channel filtering, while hierarchical deep supervision (DS) enforces consistent multiscale optimization. Experiments on LEVIR-CD, WHU-CD, and DSIFN-CD demonstrate the state-of-the-art performance with sharper boundaries and more reliable change maps. Further evaluation on the CLCD dataset verifies the robustness and generalization capability of CE-FRN.

Landslides are a common, destructive form of geological disaster that poses a threat to both infrastructure and human life. The Segment Anything Model (SAM) is a strong segmentation model, but it still has difficulty with the unclear boundaries and complex terrain that are typical of landslides. Most existing multimodal models also have difficulty in deeply fusing Digital Elevation Models (DEM) and optical images. To address these issues, we introduce MMCA-SAM, a terrain-aware multimodal segmentation model. MMCA-SAM incorporates a Cross-Attention Fusion Module (CAFM) to align RGB semantics with terrain geometry. It also incorporates Atrous Spatial Pyramid Pooling (ASPP) and a decoder with Coordinate Attention (CA) to improve the resolution of unclear boundaries. Experiments on the Bijie and Landslide4Sense datasets demonstrate that MMCA-SAM achieves better performance than existing semantic segmentation models and SOTA foundation models. Analysis also shows that topographic constraints lead to a significant improvement in landslide spatial localization accuracy. This method, aiming to obtain accurate boundary geometry, provides reliable spatial assistance for accurate earthwork estimation and damage assessment after a disaster.

Multiple network interfaces equipped with terminal nodes in heterogeneous wireless networks can be used for multi-path transmission, transmitting data in parallel to improve the throughput of network communication, as well as the stability and resilience of the connection. However, existing multipath congestion control algorithms have difficulty in dealing dynamic changing network environments and ignore the subflow coupling feature, resulting in problems such as underutilization of multipath resources and transmission fairness. To address these challenges, we propose a novel multi-path intelligent congestion control algorithm based on subflow coupling perception (MSCP), which utilizes deep reinforcement learning (DRL) techniques to improve the adaptability to dynamic network environments. The subflow coupling features are perceived through the analysis of trends in subflow transmission round-trip time. To further refine this understanding, Long Short-Term Memory (LSTM) is utilized to eliminate network noise and extract the latent temporal information inherent in subflow states. By providing a more precise estimation of network conditions, this approach enhances the ability of DRL agents to learn more effectively from their interactions with the environment. Proximal Policy Optimization (PPO) algorithm is combined with Coupled Bottleneck Bandwidth and Round-trip propagation time (Coupled BBR) multipath congestion control algorithm, where the transmission gain coefficients of each subflow is adaptively adjusted using PPO agents to control the transmission rate of each subflows, improving the algorithm’s responsiveness to dynamic network environments. Experimental results show that the proposed algorithm effectively utilizes link resources, leading to significant improvements in both transmission throughput and transmission fairness.

Given the increasing concern over anthropogenic radionuclide contamination in the China Sea, this study focuses on 239,240Pu and 237Np in sediment cores from the Pearl River Estuary (PRE). In Lingdingyang Bay's anthropogenically disturbed proximal zone, coarse sediments showed reduced 239+240Pu (36 ± 4 to 139 ± 7 Bq/m2) and 237Np (339 ± 34 to 684 ± 73 mBq/m2) inventories, while distal fine-grained deposits exhibited higher levels of 532 ± 16 Bq/m2 and 1027 ± 58 mBq/m2, respectively. The comparatively elevated 237Np inventories reflects enhanced fluvial transport of soluble 237Np relative to Pu isotopes. The associations of Pu and Np with fine particles and organic matters suggest significant roles in controlling their fate in contaminated systems. This study provides valuable insights into the sedimentary characteristics of Pu and 237Np in the PRE, significantly influenced by geographical factors and anthropogenic activities, and serves as a baseline for future nuclear safety assessments.

The 137 Cs accumulation peaks recorded in lake sediments resulted from human nuclear activities have been served as im-portant time markers and widely applied in modern sediment dating.In the lake sediment profiles collected in the Yunnan area,ex-cept for the common appeared main 137 Cs accumulation peaks of 1963,whether there were sub-peaks of 137 Cs corresponding to the year 1986 and/or 1975,the reasons for the formation of those so-called sub-peaks as well as whether they can be time markers were still controversial.In this study,Lake Qinghai in Tengchong,a natural volcanic lake located in western Yunnan,was selected as the major research object.By modeling the vertical distributions of 137 Cs in sediment cores based on the reconstructed regional at-mospheric deposition data of 137Cs,combined with the results of the 210Pbex age data and the distribution of plutonium isotopes,as well as the previous researches on other lakes in the region,the possible corresponded ages of the main accumulation peaks and sub-peaks of 137 Cs in sediment profiles and their formation reasons as well as the influencing factors were comprehensively dis-cussed,and the corresponded time markers of 137 Cs in the region were further identified.The results indicated that the 137 Cs peaks of 1963 and 1986 in sediment profiles of Lake Qinghai in Tengchong were respectively corresponded to the deposition of the global fallout from the nuclear weapon tests in 1963 and the regional fallout from the 1986 Chernobyl Nuclear Accident(CNA),with the contributions of 77％-80％and 51％-62％,respectively,showing the clear time-mark characteristics.While the 137Cs peak of 1975 in the sediment profiles was predominantly contributed by the catchment-derived 137 Cs with the proportion of 80％-88％,im-plying that this peak could not serve as a time marker corresponding to the regional fallout from the Chinese nuclear tests in 1975.This study suggested that in the Yunnan Plateau region,the maximum peak of 137Cs provides a reliable time marker in 1963,and the sub-peak contributed from the 1986 CNA can also be identified as the time marker of the year 1986 for the modern sediment dating in the region.

With the development of deep learning, the application of remote sensing (RS) image change detection (CD) has achieved essential breakthroughs. However, RS image CD faces challenges due to ill-posed problems, inherent uncertainty, and instability in the solution process. These challenges can be broadly categorized into two main issues: semantic redundancy and insufficient input information. Semantic redundancy in RS data introduces overlapping and repetitive features that complicate the CD process. Redundant and overlapping semantic content reduces the availability of clear information, introducing additional noise and making it harder to distinguish relevant changes from background variations. Insufficient input information exacerbates the ill-posed nature of RS image CD. This insufficiency often results from the loss of critical feature information during the extraction process. To address these problems, this article proposes RISNet, a fully supervised transformer-based CD method. RISNet consists of three main modules: temporal feature extraction (TFE), feature reconstruction encoder (FRE), and data perturbation regularization (DPR). First, the TFE module extracts rich spatio-temporal features from bitemporal RS images by leveraging contextual similarity to highlight change regions. Next, the FRE module employs an encoder-decoder system to analyze these features, assign change probabilities, and mask those with high change probabilities. Finally, during training, the DPR module introduces perturbations into the feature representations to enhance the model's adaptability to nonsemantic changes. This integrated approach effectively tackles the challenges of ill-posed problems. It improves model stability, demonstrating superior performance on the LEVIR, WHU, DSIFN, and S2Looking datasets compared to recent transformer-based methods.

Graph convolutional network (GCN)-based methods for hyperspectral image (HSI) classification have received more attention due to its flexibility in information aggregation. However, most existing GCN-based methods in HSI community rely on capturing fixed $K$ -hops neighbors for feature information aggregation, which ignores the inherent imbalance in class distributions and fails to achieve optimal feature smoothing through the graph convolution operator. It is unreasonable to apply a fixed $K$ -hops strategy for feature smoothing in imbalanced classes, as class regions with rich contextual information and those with poor contextual information require to capture different hops neighbors to achieve the optimal feature smoothing. To address this issue, this article proposes a novel approach called class-imbalanced graph convolution smoothing (CIGCS) for HSI classification, which achieves adaptive feature smoothing for imbalanced class regions. First, we construct a semantic block-diagonal graph structure that describes imbalanced semantic class regions by considering label connectivity and spectral Laplacian regularizer. Second, we develop the CIGCS technique to adaptively aggregate neighbor information for imbalanced class regions based on the decreasing Euclidean distance of samples within each bock-diagonal structure from the perspective of oversmoothing. The choice of adaptive neighbors can be guaranteed by a theoretical upper bound. Finally, the obtained optimal smoothed features are fed into the logistic regression to achieve good classification results. The proposed CIGCS method is evaluated on three real HSI datasets to demonstrate its superiority compared to some popular GCN-based methods.

The China Sea is faced with a heightened risk of anthropogenic radionuclide contamination, whose provenance, scavenging and migration are imperative to investigate to provide the background and nuclear safety emergency assessment. This study pioneers the measurement of anthropogenic plutonium and neptunium (239+240Pu and 237Np) concentrations and atom ratios (240Pu/239Pu and 237Np/239Pu) in sediment cores from the northern Taiwan Strait and the adjacent East China Sea using SF-ICP-MS, exploring their applications and characteristics. Typical vertical profiles confirm that Pu and 237Np serve as geochronological tools, with the 240Pu/239Pu atom ratio as a fingerprint refining the chronology. Fallout history and sedimentary environments have been reconstructed by the comprehensive application of 239+240Pu, 237Np and 210Pb chronologies. The primary sources of Pu isotopes and 237Np are global fallout and close-in fallout from the Pacific Proving Grounds (PPG). Inventories of 239+240Pu ranged from 44 ± 3 Bq/m2 to 348 ± 11 Bq/m2, with PPG contributions from 57% to 72%, while 237Np inventories varied from 58 ± 5 mBq/m2 to 137 ± 8 mBq/m2. Differences in the distribution of Pu and 237Np are attributed to their distinct behaviors and sedimentary environments. Particle-reactive Pu isotopes are predominantly preserved in sediment, whereas conservative 237Np remains mostly dissolved in water, easily re-entering seawater from sediment through resuspension processes. Higher environmental mobility also makes more downward diffusion of 237Np than Pu isotopes.

Deep learning-based medical image segmentation has made great progress over the past decades. Scholars have proposed many novel transformer-based segmentation networks to solve the problems of building long-range dependencies and global context connections in convolutional neural networks (CNNs). However, these methods usually replace the CNN-based blocks with improved transformer-based structures, which leads to the lack of local feature extraction ability, and these structures require a huge number of data for training. Moreover, those methods did not pay attention to edge information, which is essential in medical image segmentation. To address these problems, we proposed a new network structure, called P-TransUNet. This network structure combines the designed efficient P-Transformer and the fusion module, which extract distance-related long-range dependencies and local information respectively and produce the fused features. Besides, we introduced edge loss into training to focus the attention of the network on the edge of the lesion area to improve segmentation performance. Extensive experiments across four tasks of medical image segmentation demonstrated the effectiveness of P-TransUNet, and showed that our network outperforms other state-of-the-art methods.

Compressing hyperspectral images (HSIs) into compact representations under the premise of ensuring high-quality reconstruction is an essential task in HSI processing. However, existing compression methods usually encode images by smoothing due to the low-frequency information occupying a prominent component in most images. Consequently, these methods fail to capture sufficient structural information, especially in low bit rates, often causing inferior reconstruction. To address this problem, we propose here an edge-guided hyperspectral compression network, called CENet, to realize high-quality reconstruction. To enhance the structural latent representation ability, the CENet model incorporates an edge extractor neural network into the compression architecture to guide compression optimization by the edge-guided loss. We propose an interactive dual attention module to selectively learn edge features, obtain the most effective edge structure, and avoid additional edge information redundancy at the same time. In the proposed CENet, the edge-guided loss and interactive dual attention module are combined to enhance the comprehensive structure of the latent representation. Concretely, interactive dual attention makes the edge extraction network focus only on moderate boundaries rather than on all edges, which enables savings on the bit rate cost and helps achieve a strong structural representation. As a result, the reconstruction quality is significantly improved. The extensive experiments on seven HSI datasets verify that our model can effectively raise the rate–distortion performance for HSIs of any type or resolution (e.g., yielding an average peak signal-to-noise ratio (PSNR) of 30.59 dB at 0.2382 bpppb, which exceeds the baseline for Chikusei by 10.99%).

Although various methods based on convolutional neural networks have improved the performance of biomedical image segmentation to meet the precision requirements of medical imaging segmentation task, medical image segmentation methods based on deep learning still need to solve the following problems: (1) Difficulty in extracting the discriminative feature of the lesion region in medical images during the encoding process due to variable sizes and shapes; (2) difficulty in fusing spatial and semantic information of the lesion region effectively during the decoding process due to redundant information and the semantic gap. In this paper, we used the attention-based Transformer during the encoder and decoder stages to improve feature discrimination at the level of spatial detail and semantic location by its multihead-based self-attention. In conclusion, we propose an architecture called EG-TransUNet, including three modules improved by a transformer: progressive enhancement module, channel spatial attention, and semantic guidance attention. The proposed EG-TransUNet architecture allowed us to capture object variabilities with improved results on different biomedical datasets. EG-TransUNet outperformed other methods on two popular colonoscopy datasets (Kvasir-SEG and CVC-ClinicDB) by achieving 93.44% and 95.26% on mDice. Extensive experiments and visualization results demonstrate that our method advances the performance on five medical segmentation datasets with better generalization ability.

Although graph convolutional network (GCN) has achieved remarkable success in hyperspectral image (HSI) classification, most existing GCN-based approaches have failed to realize a deep network structure due to the oversmoothing problem. This problem largely limits the expression ability and feature extraction ability of GCN and hampers GCN's capacity to model long-range relationships between samples in hyperspectral (HS) scenes. Moreover, there is a lack of theoretical analysis in those works that constructed deep GCN for HSI classification to illustrate how they overcome the oversmoothing problem. Aside from this, the characteristics and complexity of HSI are often neglected when constructing deep GCN models in HSI classification. To address these problems, a novel deep graph network based on first-order smoothing is proposed for HSI classification. Specifically, a local and global topologically consistent graph is constructed to thoroughly explore the union between fine pixel information and semantic superpixel information. Subsequently, a novel propagation procedure is proposed to address the oversmoothing problem. We creatively build a residual connection to the first layer to emphasize the feature information aggregated from the first-order neighborhood, which adds node features that have not yet become indistinguishable into deep layer, and at the same time, it can be considered as a correction to the original pixels affected by spectral variation in the input graph. Finally, we demonstrate how first-order smoothing-based deep graph network (FSDGN) can slow down the convergence rate of the oversmoothing problem by analyzing the propagation of FSDGN from the standpoint of the Laplacian spectral domain. In addition, the results of experiments performed on three benchmark datasets demonstrate its superiority over other state-of-the-art methods.

沉积物粒度蕴含着区域降水强度、湖泊水位波动、沉积物物源等古环境信息.文章以二龙湾2个沉积物岩芯为研究对象,利用GMX30P-A高纯Ge同轴探测器和英国Mastersize2000型激光粒度仪对样品进行137Cs比活度和粒度测试,运用参数化EMA端元模型对沉积物样品的粒度数据集进行端元分解,探讨二龙湾近百年来沉积物质来源和沉积环境特征.二龙湾沉积物主要为水成沉积,含少量风成沉积,EM1和EM2指示水动力较弱的封闭湖泊沉积环境,EM3指示强地表径流搬运过程.坡面径流挟带的流域碎屑颗粒物是沉积物的主要物源,高空西风携带的远源粘粒级颗粒和中尺度区域风搬运的近源粉尘物质对沉积物有一定的贡献.近百年来二龙湾经历封闭湖泊沉积过程,沉积环境较为稳定,湖心沉积速率小于近岸沉积速率.1963年后沉积物中细颗粒物质含量增加,粗颗粒物质占比大幅下降,表明吉林龙湾国家级然保护区的建成对流域的水土保持有重要作用.

The recognition of fine-grained objects is crucial for future remote sensing applications, but this task is faced with the few-shot problem due to limited labeled data. In addition, the existing few-shot learning methods do not consider the unique characteristics of remote sensing objects, i.e., the complex backgrounds and the difficulty of extracting fine-grained features, leading to suboptimal performance. In this study, we developed an improved task sampling strategy for few-shot learning that optimizes the target distribution. The proposed approach incorporates broad category information, where each sample is assigned both a broad and fine category label and converts the target task distribution into a fine-grained distribution. This ensures that the model focuses on extracting fine-grained features for the corresponding broad category. We also introduce a category generation method that ensures the same number of fine-grained categories in each task to improve the model accuracy. The experimental results demonstrate that the proposed strategy outperforms the existing object recognition methods. We believe that this strategy has the potential to be applied to fine-grained few-shot object recognition, thus contributing to the development of high-precision remote sensing applications.

In recent years, advances in deep learning have greatly promoted the development of hyperspectral image (HSI) compression algorithms. However, most existing compression approaches directly rely on rate–distortion (RD) optimization without other guidance during model learning. Therefore, this brings challenges to distinguishing similar features or objects that are widely available in HSIs, especially in remote sensing scenes, since quantification in lossy compression can cause informative attribute (e.g., category) collapse and loss problems at high compression ratios. In this article, we propose a novel hyperspectral compression network via contrastive learning (HCCNet) to help generate discriminative representations and preserve informative attributes as much as possible. Specifically, we design a contrastive informative feature encoding (CIFE) to extract and organize discriminative attributes from the original HSIs by enlarging the discrimination over the learned latents in different channel indexes to relieve attribute collapses. In the case of attribute losses, we define a contrastive-invariant feature recovery (CIFR) to discover the lost attributes via contrastive feature refinement. Experiments on five different HSI datasets illustrate that the proposed HCCNet can achieve impressive compression performance, such as improvement of the peak signal-to-noise ratio (PSNR) from 28.86 dB [at 0.2284 bit per pixel per band (bpppb)] to 30.30 dB (at 0.1960 bpppb) on the Chikusei dataset.

Neptunium-237, owing to its long half-life (t1/2 = 2.14 × 106 year) and similar conservatism to 137Cs, has the potential to replace 137Cs for water mass circulation studies on decades and even longer time scales. A new method for the determination of 137Cs, 237Np, and Pu isotopes in seawater samples was proposed to solve the difficulty of 237Np analysis in seawater. The developed method includes the separation technique of ammonium phosphomolybdate (AMP) adsorption for 137Cs and anion exchange chromatography for 237Np and Pu, a measurement technique of gamma spectrometry for 137Cs and SF-ICP-MS for 237Np and Pu isotopes. 242Pu as a pseudo isotope dilution tracer for Np, the negligible chemical fractionation between 237Np and 242Pu of 1.02 ± 0.06 (k = 2) was obtained by implementing sophisticated control of the redox system and chromatographic elution optimization. The analytical results for the International Atomic Energy Agency Certified Reference Materials (IAEA-443) agreed with the reference values, showing chemical yields of 65-88%, U decontamination factor above 106 level, and improved sample throughput (5 days for 12 samples). Meanwhile, the lower method detection limits (MDLs) of 237Np, 239Pu, and 240Pu were 1.3 × 10-3, 0.065, and 0.15 μBq L-1 for 15 L seawater, respectively. Results obtained by the developed method can be used to evaluate the impact on the marine ecological system of the planned marine discharge of Fukushima decontaminated wastewater. Working toward that purpose, we are the first to report the 237Np activity concentration in Pacific Ocean seawater sampled near the station site, and we obtained the value of 0.122-0.154 μBq L-1.

In hyperspectral image (HSI) classification task, semisupervised graph convolutional network (GCN)-based methods have received increasing attention. However, two problems still need to be addressed. The first is that the initial graph structure in the GCN-based methods is not sufficiently flexible to encode the homogenous structure similarity of HSI pixels when facing the complex scenarios induced by the spatial variability. Another problem is that the input (graph structure) and output (output features) of the GCN-based methods are separated with a “single pass” procedure, which is a suboptimal problem for HSI classification because it does not flexibly optimize the graph construction with a feedback method via output features. In this article, a novel spatial–spectral unified adaptive probability GCN (SSAPGCN) method is proposed for HSI classification. First, considering the homogeneous structural similarity of the pairwise relationships of HSI pixels, this article combines the inherent spectral information and spatial coordinates to obtain the spatial–spectral adaptive probability graph (SSAPG) structure, which can capture the probabilistic connectivity between each pair of the homogeneous HSI pixels. Second, the SSAPG structure and GCN model are combined into a unified framework to a daptively learn both the graph structure and the output features simultaneously with feedback. Finally, the proposed SSAPGCN method with two layers is evaluated on four public HSI datasets to demonstrate its superiority over different classification methods in terms of two evaluation metrics, the overall accuracy (OA) and kappa coefficient (KC), especially with small training sample sizes.

Hyperspectral image (HSI) classification methods based on the graph convolutional network (GCN) have received more attention because they can handle irregular regions by graph encoding techniques. However, GCN-based HSI classification methods are highly sensitive to the quality of the graph structure. Its performance degrades in the case of underdeveloped graphs because it cannot excavate the intrinsic adjacency relationships. Thus, it is necessary to improve the quality of graph structure in GCN-based methods. In this article, a novel diversity-connected GCN (DCGCN) method is proposed to improve the quality of the graph structure for HSI classification, and its basic idea can be adopted by other GCN-based methods. First, the potential neighbors are excavated by performing topological extensions based on the given graph. The diversity of surrounding neighbors is maintained by adaptively smoothing operation via a global threshold value from Kullback–Leibler (KL) divergence to eliminate weak interclass connections caused by weakly spectral variability. Second, another key connectivity restriction is imposed on the diverse neighbors to further refine the ambiguous connections of hard samples aiming at removing strong interclass connections where the spectral information is heavily confounded. Finally, the DCGCN method is analyzed theoretically to demonstrate its low-pass filter property. The comprehensive experiments demonstrate the effectiveness of the proposed DCGCN method and the basic idea of the diversity-connected graph in terms of overall accuracy (OA), kappa coefficient (KC), and average accuracy (AA) indices.

To understand the possible influence of the Fukushima Dai-ichi Nuclear Power Plant (FDNPP) accident on the deep sea, as well as the geochemical behavior and transport of radionuclides, 134Cs, 137Cs, 239, 240Pu, 241Pu, and 237Np were measured in the abyssal sediments of the Northwest Pacific (NWP) and Central Equatorial Pacific (CEP) Ocean. Data on the characteristics of these sediments obtained after the FDNPP accident are extremely rare, especially in the NWP subtropical gyre (NPSG) region. FDNPP-derived radio-Cs (134Cs, 137Cs) arrived at the open sea floor of the NWP before 2018 but was only found in the Kuroshio-Oyashio Extension (KOE) region. No FDNPP-derived Pu was detected in the abyssal sediments of the NWP or CEP. Pu in the NWP mainly originated from global fallout and the Pacific Proving Ground (PPG) close-in fallout, except for at station WP1 (39°N in the KOE region), where an abnormal but non-FDNPP-derived Pu signal was detected. Pu in the eastern CEP sediment was less affected by the PPG close-in fallout from the Marshall Islands and was mainly derived from global fallout, with some close-in fallout from the Johnston Atoll test. The KOE region was the area most affected by PPG close-in fallout Pu via Kuroshio transport, while the lowest inventories of 239+240Pu and 237Np were found in the NPSG region due to its oligotrophic environment. The 237Np originated from the same source as Pu, and the latitudinal pattern of 237Np was consistent with that of Pu. Station SS (in the marginal sea of the NWP) contained high 237Np/239Pu atom ratios in the deeper layers of sediment and had a 237Np depth profile opposite that of the 239+240Pu profile, compared to other stations; these differences are mainly attributed to differences in the behaviors of 237Np and 239Pu.

In content-based image compression, the importance map guides the bit allocation based on its ability to represent the importance of image contents. In this paper, we improve the representational power of importance map using Squeeze-and-Excitation (SE) block, and propose multi-depth structure to reconstruct non-important channel information at low bit rates. Furthermore, Dynamic Receptive Field convolution (DRFc) is introduced to improve the ability of normal convolution to extract edge information, so as to increase the weight of edge content in the importance map and improve the reconstruction quality of edge regions. Results indicate that our proposed method can extract an importance map with clear edges and fewer artifacts so as to provide obvious advantages for bit rate allocation in content-based image compression. Compared with typical compression methods, our proposed method can greatly improve the performance of Peak Signal-to-Noise Ratio (PSNR), structural similarity (SSIM) and spectral angle (SAM) on three public datasets, and can produce a much better visual result with sharp edges and fewer artifacts. As a result, our proposed method reduces the SAM by 42.8% compared to the recently SOTA method to achieve the same low bpp (0.25) on the KAIST dataset.