Deep learning has shown great promise in hyperspectral unmixing (HU), especially the unmixing methods based on autoencoder (AE) networks, which are the most prevalent these days. Since most spectral mixing problems are nonlinear and cannot effectively utilize global context information, these methods have limited generalization ability under different ground features and scenarios. To address these limitations, an enhanced network based on spatial-spectral information attention is proposed. A two-channel attention mechanism is embedded within the convolutional AE to acquire the feature dependencies. The spatial information extraction uses the dynamic large kernel block (DLK) to obtain the global spatial attention of the image. The DLK module uses multiple large kernels with different kernel sizes and dilation rates to capture multiscale features, and the spectral information extraction uses coordinate attention (CA) to capture spectral correlation information. This can improve the quality of the endmember spectra and abundance maps. On real and synthetic data, this model is compared with several advanced unmixing methods, and the results show the effectiveness of this method.

Stance detection, which focuses on users' deep attitudes, is an important way to understand the online public opinion. This paper presents an overview of stance detection. First, we present a general framework for stance detection, and the main steps of the framework are introduced in detail. The state-of-the-art stance detection methods are categorized into three classes: feature-based methods, deep learning-based methods, and ensemble learning-based methods. Moreover, the advantages and limitations of the existing methods are analyzed. The survey findings show that hybrid-neural network-based methods are superior to the other methods. In addition, existing methods still need to pay more attention to the sentiment information, user-interaction, and attempt to merge more external knowledge to help improve the effect of stance detection.

When several Web services with simple functions need to be combined to provide more complex functions, how to choose from a large number of Web services with the same functions but different quality of service is a QoS-based service composition problem. Currently, there are many classical methods and reinforcement learning methods applied to the QoS-based service composition problem. However, these methods require long computation time. We address three challenges in building an end-to-end supervised learning framework. 1) The number of Web services composing different composite services varies. 2) The topological relationships among Web services are difficult to express and difficult to integrate into neural networks. 3) The number of Web services providing each sub-function in composite services varies. Finally, we propose DeepQSC, a deep supervised learning framework based on graph convolutional networks and attention mechanisms. The framework can form high QoS composite services with limited computation time. We conducted experiments on a real-world dataset. The experiments show that DeepQSC has a significant advantage over six current state-of-the-art algorithms.

Motivated by the success of Bitcoin, lots of cryptocurrencies have been created, the majority of which were implemented as smart contracts running on Ethereum and called tokens. To regulate the interaction between these tokens and users as well as third-party tools (e.g., wallets, exchange markets, etc.), several standards have been proposed for the implementation of token contracts. Although existing tokens involve lots of money, little is known whether or not their behaviors are consistent with the standards. Inconsistent behaviors can lead to user confusion and financial loss, because users/third-party tools interact with token contracts by invoking standard interfaces and listening to standard events. In this work, we take the first step to investigate such inconsistent token behaviors with regard to ERC-20, the most popular token standard. We propose a novel approach to automatically detect such inconsistency by contrasting the behaviors derived from three different sources, including the manipulations of core data structures recording the token holders and their shares, the actions indicated by standard interfaces, and the behaviors suggested by standard events. We implement our approach in a new tool named TokenScope and use it to inspect all transactions sent to the deployed tokens. We detected 3,259,001 transactions that trigger inconsistent behaviors, and these behaviors resulted from 7,472 tokens. By manually examining all (2,353) open-source tokens having inconsistent behaviors, we found that the precision of TokenScope is above 99.9%. Moreover, we revealed 11 major reasons behind the inconsistency, e.g., flawed tokens, standard methods missing, lack of standard events, etc. In particular, we discovered 50 unreported flawed tokens.

With the rapid growth of multimedia message exchange and digital communication, multimedia big data has become a research hotspot in various fields. The storage and transmission of multimedia big data have high requirements for security. Images, covering the highest proportion of multimedia data, should be processed and transmitted with high security. Compressive sensing (CS) has a beneficial property for the encryption that the image can be recovered with fewer samples than conventional approaches use. In recent years, CS has been studied not only to reduce the resource requirements for signal acquisition but also to ensure the security of data. It is still an open challenge to improve security and enhance the quality of the decrypted image simultaneously using the key with small size. In this article, a CS-based encryption method is presented that associates the quantization with random measurement permutation. An enormous number of experiments have been conducted on both standard test images and face images chosen from the big database LFW. Experimental results show that our proposal has dramatic improvements on ensuring the security, enhancing the quality of the decrypted image, and raising the efficiency. Additionally, this proposal remarkably reduces storage and transmission resources. Accordingly, this encryption scheme can be applied to ensure the security of multimedia transmission.

Combining historical white and black aerial photo with more recent LiDAR and high resolution imagery, this research mapped wetland with high accuracy from 1930s to 2010s in object-based image analysis approach (OBIA). This research shows good potential in combining grey level information with OBIA method to map accurate historical wetland. We found that there were more small wetlands in 1938, but more large wetlands in 1978. In 2011, there were similar amount of small wetlands with 1938. Though 2011 had fewer large wetlands than 1978, 2011 year's individual large wetland had much larger area than those in 1978 and thus contributed to much larger total wetland area in 2011 than in 1978. We found significantly increasing precipitation and decreasing temperature over the time series, also drought in 1938, and this may explain wetland change well.