Most existing music recommendation systems struggle to perceive users’ implicit emotional states and fail to adapt dynamically to evolving preferences in emotionally rich, context-sensitive scenarios. To address this limitation, we propose an emotion-aware conversational music recommender built on a multiagent system. The system incorporates specialized agents for emotion recognition, semantic intent analysis, and contextual understanding. It distinguishes between explicit emotions, which are directly expressed by the user (e.g., “I feel anxious”), and implicit emotions inferred from contextual cues such as time, environment, or behaviors the user may not be fully aware of. A dual-memory mechanism models long-term musical preferences using a linear decay function and captures short-term, emotion-driven preferences using exponential decay. To enrich music content understanding, multisource information fusion combines streaming platform suggestions with rich metadata from external repositories. The system employs a large language model (LLM) to conduct multiturn dialogues and generate personalized, explainable recommendations. Experimental results show that the proposed approach significantly outperforms existing platforms (e.g., Spotify and Last.fm) in recommendation accuracy, ranking performance, and Hit Ratio@K. These findings underscore the effectiveness of integrating multiagent collaboration, emotion modeling, memory-augmented user profiling, and multisource data fusion for adaptive, user-centric music recommendation.

Language models are trained to predict the next words for given content. The question arises: do large language models (LLMs), which have achieved technological breakthroughs in natural language generation, possess decision-making abilities as humans? The present article lets a series of LLMs do cognitive psychology experiments to explore this hypothesis. More specifically, we first identified a choice problem set and then conducted experiments following the certainty equivalent (CE) paradigm. We find that much of LLMs’ behaviors are consistent with humans: the fourfold pattern of risk attitudes can also be observed across the most LLMs subjects. Particularly, LLMs are risk-seeking and risk-averse for gains with low and high probabilities, LLMs are risk-averse and risk-seeking for losses with low and high probabilities. Furthermore, LLMs subjects exhibit greater rationality in losses than in gains, and less rationality during gains compared to losses. The exponential values of the value function and weighting function, fitted based on the data from LLMs subjects, satisfy the requirements of cumulative prospect theory (CPT). Taken together, these results will further enhance the theoretical foundations to understand the LLMs decision-making capabilities and provide novel tools from cognitive psychology for improving the interpretability of LLMs.

Recent advances in speech-driven facial animation have attracted significant interest across computer graphics, human–computer interaction systems, and immersive virtual reality applications. However, existing methods remain constrained by dependencies on specific reference videos or proprietary face mesh structures, limiting their applicability across diverse production pipelines and reducing compatibility with industry-standard animation workflows. To overcome these fundamental limitations in generalization and deployment flexibility, we propose Speech2Blend—an end-to-end hybrid convolutional-recurrent network that directly learns nonlinear speech-to-blendshape parameter mappings. This novel approach enables markerless speech-driven facial animation generation without restrictive inputs like video references or specialized facial rigs. Trained on the largest available digital human dataset (BEAT) and rigorously evaluated using three benchmark datasets with photorealistic visualization tools, Speech2Blend achieves state-of-the-art performance. It delivers superior audio-visual synchronization through learned temporal dynamics and reduces lip vertex error by 30% compared to existing baseline methods. These advances significantly lower production costs for virtual human speech animation while enabling cross-platform compatibility with common game engines and animation software.

ABSTRACT Driven by the rise of the internet, recent years have witnessed the gradual manifestation of commercial values of online reviews. In movie industry, sentiment analysis serves as the foundation for mining user preferences among diverse and multi‐layered audiences, providing insight into the market value of movies. As a representative task, aspect‐based sentiment analysis (ABSA) aims to analyse and extract fine‐grained sentiment elements and their relations in terms of discussed aspects. Relevant studies, particularly in the realm of deep learning research, face challenges due to insufficient annotated data. To alleviate this problem, we propose a guideline for fine‐grained sentiment annotations that defines aspect categories, describes the method for annotating aspect sentiment triplets, either simple or complex and designs a scheme to represent hierarchical labels. Based on this, an ABSA dataset tailored for the movie domain is curated by annotating on 1100 Chinese short reviews acquired from Douban. Applicability of both the annotation guideline and curated data is evaluated through inter‐annotator consistency and self‐consistency checks, and domain adaptation assessment of e‐commerce and healthcare cases. Predictive performance of machine learning models on this dataset shed light on possible applications in more fine‐grained sentiment analysis in the movie domain, for example, figuring out the aspects from which to stimulate viewership and influence public opinions, thereby providing substantial support for the movie's box office performance. Finally, we extended our fine‐grained sentiment annotation guideline to the e‐commerce and healthcare. Through empirical experimentation, we demonstrated the universality of these guideline across diverse domains.

ABSTRACT The widespread use of Internet has accelerated the explosive growth of data, which in turn leads to information overload and information confusion. This makes it difficult for us to communicate effectively in social groups, thereby intensifying the demands for emotional companionship. Therefore, we propose a novel social group chatting framework based on Large Language Model (LLM) powered multiple autonomous agents collaboration in this article. Specifically, BERTopic is used to extract topics from history chatting content for each social group everyday, and then multiple topics tracking is realised through multi‐level association by adaptive time sliding‐window mechanism and optimal matching. Furthermore, we use topic tracking architecture and prompts to design and implement an AI Chatbot system with different characters that can conduct natural language conversations with users in online social group. LLM, as the controller and coordinator of the whole AI Chatbot for sub‐tasks, allows different AI Agents to autonomously decide whether to participate in current topic, how to generate response, and whether to propose a new topic. Each AI Agent has their own multi‐store memory system based on the Atkinson‐Shiffrin model. Finally, we construct a verification environment based on online game that is consistent with real society. Subjective and objective evaluation methods were deployed to perform qualitative and quantitative analyses to demonstrate the performance of our AI Chatbot system.

Robust and high-precision vehicle positioning information is crucial for Internet of Things (IoT) applications like autonomous driving and intelligent transportation. While global navigation satellite system (GNSS)/inertial navigation system (INS) integration ensures reliable positioning in open areas, signal interruptions in urban environments cause rapidly accumulating INS errors. Deep neural networks (DNNs) are integrated into the Kalman filter (KF) framework to mitigate INS errors through learning-based data fusion. If the DNN provides unreliable pseudo-measurement information, the positioning results may deteriorate. To enhance DNN estimation reliability, this study introduces a 1-D convolution-based spatiotemporal attention mechanism for implicit modeling of temporal dependencies and spatial correlations in sensor data. This mechanism is utilized to construct a multitask odometry network (MT-ONet) for accurate forward velocity and motion state estimation of vehicles. Building upon this, an adaptive fusion vehicle positioning algorithm is proposed. This algorithm combines the MT-ONet and an improved adaptive KF (AKF) to dynamically adjust measurement noise to enhance robustness. Experimental results indicate that MT-ONet offers high estimation accuracy and low computational complexity, making it suitable for deployment on resource-constrained IoT devices. During simulated 180-s GNSS outages, the proposed method decreases horizontal root mean square (RMS) and maximum errors by 14.47% and 16.09% compared to hardware odometer-assisted scheme, and by 33.88% and 40.28% versus the nonholonomic constraint (NHC)-assisted scheme.

The high volatility, seasonality and complex trading environment of agricultural futures markets present significant challenges to the dynamic adaptability of algorithmic trading systems. Traditional methods rely on fixed-scale image analysis and lack adaptability, making them unsuitable for varying volatility conditions. Therefore, this study proposes an AI agent for agricultural futures trading decisions. First, an agricultural futures adaptive volatility rate serves as a computational tool to dynamically identify high-volatility regions and generate finer-scale candlestick charts. Second, the agent uses Vision-Language Large Model with robust image comprehension to analyze the multiscale candlestick charts and extract key features such as trend direction and technical patterns. Subsequently, a Large Language Model with advanced natural language understanding and logical reasoning serves as the “decision-making brain”, evaluating market trends and making buy/sell decisions. Finally, the agent refines its decision logic through a multimodal feedback mechanism that combines numerical and textual information from ongoing interactions with the environment, thereby enhancing system adaptability and robustness. Experimental results indicate that this framework significantly improves the accuracy, stability, and risk control of trading strategies, offering valuable insights for human decision-making. In addition, it demonstrates potential applicability in other financial markets such as stock indices, energy, and major commodities, providing an innovative solution for intelligent trading in complex market conditions.

Social fintech envelopes social networks within financial concepts and tokenization, and its complexity requires innovative technological solutions and theories to meet user demands and achieve sustainability. The first thing is the mapping principle between physical real society and virtual digital world. Therefore, this article uses online gameNova Empireas a case study, aiming at using multiplex networks to understand the physical–digital structural consistency for social fintech sustainable development. Specifically, we first proposed an eight-layer multiplex network to model the complex gaming behaviors for players. Furthermore, we analyze the structural properties and social balance of the social networks. Particularly, the layers in multiplex networks composed of positive behaviors has higher reciprocity than the layers composed of negative behaviors, the out-degree distributions of nodes in the layers composed of negative behaviors basically conform to the power-law distribution, and the small-world phenomenon is also common in virtual game. The experimental results prove the structural consistency between physical and digital. Finally, new solutions for social fintech and mobile Internet industry are proposed based on the mapping principle, which will provide technical supports for the realization of sustainable development and social responsibility of social fintech.

Remote Sensing Image Generation (RSIG) offers a viable solution to the high data collection costs by facilitating the generation of large datasets. However, it is often hindered by the complexity of tasks and the diverse requirements of generated images. This paper presents the RS-Agent system, a novel approach that harnesses the capabilities of Large Language Models (LLMs) within an innovative agent solution, effectively addressing these issues. The system comprises a Task Agent, a Prompt Agent, and a LoRA Agent, each performing crucial roles in task decomposition, prompt generation, and scene-specific fine-tuning, respectively. Anchored by Diffusion models and employing natural language dialogues for interaction, the system aligns closely with user intent and produces high-quality results. Experimental results demonstrate the efficacy of the RS-Agent system in managing diverse RSIG tasks, adapting to various input forms, and generating high-quality results.

As a common phenomenon that often appears on social platforms, news sites, and community forums, topics have played an irreplaceable role in public opinion and social governance. Meanwhile, people's daily lives are increasingly dependent on the breeding, transformation, and attenuation of hot topics. This article aims to discuss the problem about topic duration, that is, what are the principle factors that affect topic duration? Why do some topics survive longer and even generate subtopics, while other topics disappear rapidly? To answer these questions, we innovatively use 104 121 alliance chat content inNova Empire IIfrom July 2023 to December 2023 as a case study. Dynamic topics trajectories are first obtained from a novel multilevel association model. Then, a potential factors system based on the dimensions of topic properties, topic users, and social network is established to quantitatively evaluate the influence for different factors. Experimental results from a robust statistical analysis framework demonstrate that higher topic discussion intensity, more content from opinion leader, faster information diffusion, and closer intertopic correlations will significantly improve the topic duration. Finally, a series of strategies are proposed to promote the design of social system applications from the perspectives of online social group.

With the transformation of human life into virtual worlds based on the reality, current researches of smart cities and sustainability should integrate the factors of virtual social behaviors. Virtual games are an ideal domain and tools for exploring social science. Therefore, our work innovatively using Nova Empire as a case study to reveal the social group long-term survival via complexity analysis of behavior graph structural properties for building smart cities. Specifically, behavioral data of 101424 players and 5324 alliances from September 2021 to February 2022 are used. Our observations show that the phenomenon of “gap of wealth” in real society also exists in the virtual game. Meanwhile, the player online time is significantly positive related to alliance survival time, which is the foundation of our work. Then, correlation and regression analysis are performed to understand the significance of different structural properties on alliance survival time. Our original findings demonstrate that larger alliance, more subgroups, balanced player distribution, and frequent behavioral interactions will promote long-term survival of the alliance. Small subgroup and a relaxed social environment can improve player online time. Furthermore, we transfer the conclusions from virtual game to real society based on the mapping principle. Finally, new virtual tools and solutions for policymakers to improve smart cities and sustainable society ecosystem, and operation strategies for game designers to improve players retention rate are proposed.

With the constant fusion of virtual and reality, a new vision of human beings has emerged—Metaverse. At present, the development of metaverse is still in its infancy, and although the industry has put feverish investment into it, there are still many problems that need to be discussed in academia. The first thing is user engagement. Metaverse relies heavily on massive online users to realize its social value. In other words, user engagement is the foundation of the metaverse ecosystem. Therefore, our research uses Nova Empire as a case study, aiming at understanding metaverse engagement via complexity analysis of social patterns and reward mechanisms. Specifically, the behavioral data of 46 954 players in September 2021 are used for analysis. Our observations show that social behavior is not the main factor for user engagement. Then, we perform a correlation analysis of trigger times for different gaming behaviors to verify the above observations. The results prove that the main factors affecting user engagement are game tasks and reward mechanisms in the early stage and social gaming behaviors environment with alliances in the middle and late stages. Finally, we discuss the implications of our findings for the design of future games and propose operation strategies on user engagement to improve the metaverse ecosystem and other online social space.

Human mesh recovery from one single image has achieved rapid progress recently, but many methods suffer from the image appearance overfitting since the training data are collected along with accurate 3D annotations in controlled settings of monotonous backgrounds or simple clothes. Some methods regress human mesh vertices from poses to tackle the above problem. However the mesh topologies have not been well exploited, and artifacts are often generated. In this paper, we aim to find an efficient low-cost solution to human mesh reconstruction. To this end, we propose a Progressive Quadric Graph Convolutional Network (PQ-GCN), and design a simple and fast method for 3D human mesh recovery from a single image in the wild. Specifically, we apply quadric-based surface simplification to human meshes and design a progressive graph convolution network, accompanied by mesh feature up-sampling, to deal with the mesh topologies. We carry out a series of studies to validate our method. The results prove that our method achieves superior performance on a challenging in-the-wild dataset, while using 66% fewer parameters than the existing method, Pose2Mesh. Artifacts have also been eliminated and better visual quality has been obtained without any further post-processing and model fitting. Besides, the recovery can be stopped at an earlier stage by adding a decoder head. Consequently, the computational complexity can be reduced greatly.

The characteristics of sea clutter in real marine environments in different sea areas play a vital role in military industry, such as radar detection, remote sensing, SAR imaging, and situational awareness. In this article, a deep neural network (DNN) sea clutter model is proposed based on the sea clutter big data under the real marine environment to study the characteristics of sea clutter and parameter inversion. Based on the ERA-Interim reanalysis (2015–2017), a database of marine environmental elements in China's offshore waters was established, and a spatiotemporal prediction model for marine environmental elements was proposed to improve missing values. Considering the scattering mechanism of sea surface at different scales comprehensively, a large database of sea clutter time series of multiscale real surface is established and comparison with the experimental data. Aiming at the coastal waters of China, the long short-term memory model and DNN are used to establish the correlation model between marine environmental elements and sea clutter characteristics, and the prediction and parameter inversion of sea clutter characteristics based on sea clutter big data are studied. The results show that the coefficients of determination of the predicted fitted curves for the HH and VV polarization amplitudes reach 0.9249 and 0.8872, respectively. The inversion of wave heights in different sea areas is the lowest in the South China Sea (accuracy rate of 78%) and the highest in the East China Sea (accuracy rate is 90%). The results of this article can help improve the ability of sea surface remote sensing and sea clutter suppression.

Visual relationship detection (VRD) is one newly developed computer vision task, aiming to recognize relations or interactions between objects in an image. It is a further learning task after object recognition, and is important for fully understanding images even the visual world. It has numerous applications, such as image retrieval, machine vision in robotics, visual question answer (VQA), and visual reasoning. However, this problem is difficult since relationships are not definite, and the number of possible relations is much larger than objects. So the complete annotation for visual relationships is much more difficult, making this task hard to learn. Many approaches have been proposed to tackle this problem especially with the development of deep neural networks in recent years. In this survey, we first introduce the background of visual relations. Then, we present categorization and frameworks of deep learning models for visual relationship detection. The high-level applications, benchmark datasets, as well as empirical analysis are also introduced for comprehensive understanding of this task.

Summary The northwest region in China located at arid and semiarid areas, atmospheric precipitation converted by the cloud is an important part of water resources, and if we fully utilize cloud for cloud‐water conversion to alleviate the scarcity of water, thereby it is particularly important to analyze the macroscopic characteristics and the changing trends of clouds in the northwest region. In this paper, the 2016 Level 1 data of Fengyun Satellite has been calibrated, corrected, and processed, using the improved multi‐spectral thresholding method to calculate cloud coverage and cloud classification data. The optical thickness inversion uses the SBDART (Santa Barbara DISORT Atmospheric Radiative Transfer) radiation transmission mode to establish a radiation look‐up table with optical thickness as a function variable under different conditions of observation geometry, underlying surface type, and atmospheric environment. The results show that the coverage of clouds in the northwest region accounts for about 45%, and the cloud coverage changes with the seasons. The classification of clouds mainly consists of high clouds, low clouds, and cumulonimbus. The optical thickness of clouds is largely distributed between 10 and 25.

With the development of remote sensing technology, spatial and spectral resolutions of hyperspectral images have become increasingly dense. In order to overcome difficulties in the storage, transmission, and manipulation of hyperspectral images, an effective compression algorithm is requisite. The clustered differential pulse code modulation (C-DPCM), which is a prediction-based hyperspectral image lossless compression algorithm, can achieve a relatively high compression ratio, but its efficiency still requires improvement. This paper presents a parallel implementation of the C-DPCM algorithm on graphics processing units (GPUs) with the compute unified device architecture, which is a parallel computing platform and programming model developed by NVIDIA. Three optimization strategies are utilized to implement the C-DPCM algorithm in parallel, including a version that uses shared memory and registers, a version that employs multistream, and a version that uses multi-GPU. In addition, we studied how to assign all classes to each GPU to minimize the processing time. Finally, we reduced the compression time from approximately half an hour to an hour to several seconds, with almost no loss in accuracy.

An efficient facet-based composite scattering model (FBCSM) is developed for calculating the timeevolving electromagnetic (EM) scattering field (TESF) to study the normalized radar cross section and Doppler spectrum characteristics from dynamic sea surfaces. The dynamic sea surface comprises two-scale profiles: smallscale capillary ripples modulated by large-scale gravity waves, which are modeled by millions of small facets. In microwave bands, two scattering mechanisms, quasi-specular scattering with respect to gravity waves and Bragg scattering with respect to ripples, are taken into account in the FBCSM for computation of the time-evolving EM scattering field under diverse polarizations. However, it may be very time-consuming and difficult to calculate the TESF due to the high resolution and dynamic complexity of the large dynamic sea surface. In this paper, the NVIDIA Tesla K80 graphics processing unit (GPU) with the compute unified device architecture is utilized to improve the computational performance of the TESF. The whole GPU-based TESF computation includes the optimal use of temporary variables, shared memory, constant memory and register, fastmath compiler options, asynchronous data transfer, and the most suitable block size and number of registers. By utilizing the proposed five improvement strategies, a significant speedup of 1200× can be achieved for computation of TESF from large dynamic sea surfaces for microwave bands compared with the single-threaded C program executed on the Intel(R) Core(TM) i5-3450 CPU.

The study of sea clutter reflectivity plays an important role in radar performance evaluations in the military industry. The industrial bodies are trying to apply the sea clutter intelligent processing technology to the radar system with the form of Internet of Things and Industry 4.0. Many sea clutter reflectivity models that have been proposed are difficult to fully adapt to the surrounding seas and different radar systems in China. This article proposes a model named multi-source input neural network (MSINN) sea clutter model using sea clutter collected by ultra high frequency (UHF) radar. In order to prepare sea clutter reflectivity data for training MSINN, the radar continuously collects sea clutter containing various disturbances. In the face of the challenges of preprocessing and storage of measured sea clutter big data, this article proposes a sea clutter preprocessing scheme based on yolov3-tiny model. Experimental results show that the average detection precision of test sea clutter Range-Pulse (RP) images is 75.3% and the effective region extraction time of sea clutter RP image is 0.003642 s, which can meet the requirement of real-time detection and data requirement of predicting sea clutter reflectivity based on MSINN. Compared with the traditional empirical model, the average prediction error of sea clutter reflectivity based on MSINN is 1.82 dB, which improves the prediction accuracy and is more suitable for the Yellow Sea in China.

Based on the fuzzy suppressed c-means clustering algorithm, a new method is developed for extracting auroral oval regions from images acquired by the Ultraviolet Imager aboard the POLAR satellite. Compared with different variations of fuzzy c-means clustering methods, suppressed fuzzy c-means clustering was proposed with the intention of improving convergence rate by modifying membership values, which is more suitable for studying auroral behavior over time with considering a series of images. However, traditional suppressed c-means clustering algorithms employ the same suppressed parameter for modifying fuzzy membership degrees of all pixels, ignoring the fact that image characteristics varies from one auroral oval images to another. In this paper, the technique parameters which is set beforehand will be automatically selected according to the intrinsic characteristic of each auroral oval image. Moreover, corresponding operations are devised for modifying membership values of different pixels according to their real needs, which makes it clear to decide whether to proceed with further determination or just make decision on the basis of already obtained analysis results. Experimental results on auroral oval images acquired from an online database collected by NASA Polar satellite's Ultraviolet Imager indicate that the proposed method extracts more accurate auroral oval regions than traditional suppressed c-means clustering method in most cases.

With the growth in the consumer electronics industry, it is vital to develop an algorithm for ultrahigh definition products that is more effective and has lower time complexity. Image interpolation, which is based on an autoregressive model, has achieved significant improvements compared with the traditional algorithm with respect to image reconstruction, including a better peak signal-to-noise ratio (PSNR) and improved subjective visual quality of the reconstructed image. However, the time-consuming computation involved has become a bottleneck in those autoregressive algorithms. Because of the high time cost, image autoregressive-based interpolation algorithms are rarely used in industry for actual production. In this study, in order to meet the requirements of real-time reconstruction, we use diverse compute unified device architecture (CUDA) optimization strategies to make full use of the graphics processing unit (GPU) (NVIDIA Tesla K80), including a shared memory and register and multi-GPU optimization. To be more suitable for the GPU-parallel optimization, we modify the training window to obtain a more concise matrix operation. Experimental results show that, while maintaining a high PSNR and subjective visual quality and taking into account the I/O transfer time, our algorithm achieves a high speedup of 147.3 times for a Lena image and 174.8 times for a 720p video, compared to the original single-threaded C CPU code with -O2 compiling optimization.

The proportion of an aurora region in a field of view is an important index to measure the magnetic stress stored in the magnetosphere. Detecting the aurora region is a necessary step to obtain the index. Intensity inhomogeneity, a characteristic of overlaps between the ranges of intensities in segmented regions, has become a challenging issue in the field of auroral oval segmentation. Classical auroral segmentation methods can reasonably detect auroral ovals in clean images. The segmentation quality of these methods deteriorates when auroral oval pixel intensities are not distinct from the background. To reduce the negative influence of intensity inhomogeneity in auroral oval segmentation, a gradually evolved active contour model employing the narrow-band technique instead of using a full region computation is designed. In such case, only the region near auroral oval boundaries has to evolve in each iteration, thus enabling the contour to evolve gradually and saving computational resources. Experimental results demonstrate that the proposed method detects more accurate auroral oval regions than traditional methods in terms of human visual perception and segmentation accuracy.

To the problem that random placement in distributed file system leads to the low utilization of servers, this article models according to the characteristics of remote sensing image data blocks, clusters data by setting storage centre with access frequency so that the adjacent remote sensing image data blocks in spatial position are near each other in physical storage as well. It promotes the response speed of the system, places data blocks according to data block groups, reshuffle the non-grouped data blocks at the system low load and turns off dispensable datanodes to achieve energy saving. The experiments suggest that in the inquiry of remote sensing image data, it is more efficient to place data blocks according to their own characteristics than random placement. Compared with common dynamic data placement strategies, this strategy performs better in energy saving when the system is at moderate load.

Demosaicking is a digital image process to reconstruct full color digital images from incomplete color samples from an image sensor. It is an unavoidable process for many devices incorporating camera sensor (e.g., mobile phones, tablet, and so on). In this paper, we introduce a new demosaicking algorithm based on polynomial interpolation-based demosaicking. Our method makes three contributions: calculation of error predictors, edge classification based on color differences, and a refinement stage using a weighted sum strategy. Our new predictors are generated on the basis of on the polynomial interpolation, and can be used as a sound alternative to other predictors obtained by bilinear or Laplacian interpolation. In this paper, we show how our predictors can be combined according to the proposed edge classifier. After populating three color channels, a refinement stage is applied to enhance the image quality and reduce demosaicking artifacts. Our experimental results show that the proposed method substantially improves over the existing demosaicking methods in terms of objective performance (CPSNR, S-CIELAB ΔE*, and FSIM), and visual performance.

A high-order clustered differential pulse code modulation method with removal of local spectral outliers (C-DPCM-RLSO) is proposed for the lossless compression of hyperspectral images. By adaptively removing the local spectral outliers, the C-DPCM-RLSO method improves the prediction accuracy of the high-order regression predictor and reduces the residuals between the predicted and the original images. The experiment on a set of the NASA Airborne Visible Infrared Imaging Spectrometer (AVIRIS) test images show that the C-DPCM-RLSO method has a comparable average compression gain but a much reduced execution time as compared with the previous lossless methods.

Edge-directed interpolation is widely used to enhance visual performance of remote sensing image. Compared with traditional bi-cubic interpolation and bilinear interpolation, a great number of matrix operations will appear as it is getting better visual performance. CUDA (Compute Unified Device Architecture) offers tremendous performance in many high-performance computing areas. Edge-directed interpolation can be mapped to this architecture (CUDA) readily. However, parallel schemes based on CUDA are generally decomposed into coarse-grained tasks, which is suitable for thread blocks. In this paper, a parallel approach of fine-grained edge-directed interpolation is proposed. Based on CUDA, the process of parallel interpolation for one missing pixel is assigned to 4*4 threads for the reason that majority of matrix operations are related to 4*4 matrix. This task division strategy minimizes resource pressure of thread-blocks. Our calculating scheme is expressed in terms of increasing parallelism that is efficiently implemented on the GPU. By employing one NVIDIA GTX480 GPU and one NVIDIA GTX590 GPU in the case with asynchronous I/O transfer, our GPU optimization efforts on fine-grained edge-directed interpolation scheme finally achieve a speedup of 69.8x with respect to its CPU counterpart C code running on one CPU core of Intel core(TM) i7-920.

Graphics Processing Unit (GPU) is used to accelerate the calculation of scattering of complex target from background radiation in infrared spectrum. Compute Unified Device Architecture (CUDA), OpenACC, and Hybrid Multicore Parallel Programming (OpenHMPP) implementations are presented. In all our implementation, scattering of background radiation in different directions are calculated in parallel. A personal desktop with 2 NVIDIA GTX GeForce 590 with an Intel i7 CPU is used in our experiment. In CUDA, by using shared memory to buffer the background radiation and BRDF parameters and tuning the grid organization, we achieve a speedup of 197x. OpenACC implementation is realized by inserting the parallel loop construct with reduction clause before the loop in original serial code. By utilization of data clause and tuning number of gangs used, a speedup of 158.9x is obtained. In OpenHMPP implementation, the loop iterating over incident direction of original code is transformed to the codelet function and we achieve a speedup of 160.7x. Our effort makes the calculation of complex target in real time possible.

Scattering characteristic of complex targets from sky and ground background radiance plays an important role in engineering fields. Firstly, a 5-parameter BRDF (Bidirectional Reflectance Distributional Function) model is introduced. Then MODTRAN is used to calculate the background radiance in infrared spectrum of 3-5 um and 8-12um bands. Considering the background radiance comes from all directions of space in large numbers of different bands, there will be multiple loops in the computation thus it's quite time-consuming. Thanks to the NVIDIA CUDA (Compute Unified Device Architecture), programing GPU does not require as much knowledge about graphics card and complex programing interfaces as before. On the basis of CUDA, a parallel implementation is presented and to get a higher speedup the code is optimized to reduce the access latency as much as possible by using the shared and constant memory on GPU. The implementation is test on an NVIDIA GTX GeForce 480 and 2.79 GHz Intel i7 CPU. Compared to the CPU implementation, we achieve a peak speedup of 308 and results showing the efficiency of the parallelism and optimization.

We report here a backscattering model of average signal power function (SPF) for laser radar 3D range imagery obtained by arrays of detectors for arbitrary coarse targets. The model relates the average power seen by the receiver with laser pulse, target shape, optical scattering properties of surface material, incidence angle and other factors. The optical scattering property of the material is characterized by bidirectional reflectance distribution function (BRDF). The model can be used for demonstration of 3D laser radar system and can also be used to generate library of model data sets for automatic target recognition (ATR). Finally, Compute Unified Device Architecture (CUDA) is introduced for parallelizing these algorithms. The acceleration reaches 56 times speedup on single Fermi-generation NVIDIA GTX 480 as compared to the traditional CPU version of code on Intel i7 930 CPU.

Recovery algorithms play a key role in compressive sampling (CS). Currently, a popular recovery algorithm for CS is the orthogonal matching pursuit (OMP), which possesses the merits of low complexity and good recovery quality. Considering that the OMP involves massive matrix/vector operations, it is very suited to being implemented in parallel on graphics processing unit (GPU). In this paper, we first analyze the complexity of each module in the OMP and point out the bottlenecks of the OMP lie in the projection module and the least-squares module. To speedup the projection module, Fujimoto's matrix-vector multiplication algorithm is adopted. To speedup the least-squares module, the matrix-inverse-update algorithm is adopted. Experimental results show that +40x speedup is achieved by our implementation of OMP on GTX480 GPU over on Intel(R) Core(TM) i7 CPU. Since the projection module occupies more than 2/3 of the total run time, we are looking for a faster matrix-vector multiplication algorithm.

The edge-directed interpolation scheme is a non-iterative, orientation-adaptive method to enhance image resolution with better visual effect than conventional interpolation methods. It interpolates the missing pixels based on the covariance of a high-resolution image estimated from the covariance of the low-resolution image. In spite of the impressive performance, the computational complexity of covariance-based adaptation is significantly higher than that of the conventional linear interpolation algorithms. In this paper, we propose a GPU-based massively parallel version of the edge-directed interpolation scheme. A speedup of 61.7x can be achieved with respect to its single-threaded CPU counterpart in the host computer.

In this letter, we propose a shape-adaptive (SA) reversible integer lapped transform (SA-RLT) method. The new method can deal with arbitrarily shaped image areas while guaranteeing completely reversible integer-to-integer transform. Based on SA-RLT and object-based set partitioned embedded block coder, a new region-of-interest (ROI) compression scheme is designed for 2-D remote sensing images. Numerical experiments reveal that SA-RLT performs better than integer SA discrete wavelet transform, and the new ROI compression scheme performs comparably even better than the JPEG2000-ROI scheme. Advantages in hardware implementation have been preserved by SA-RLT, such as parallel processing and low memory requirement.

A multiplierless low complexity reversible integer Karhunen-Loe¿ve transform (Low-RKLT) is proposed based on matrix factorization. Conventional methods based on KLT suffer from high computational complexity and unability of applying in lossless medical image compression. To solve the two problems, multiplierless Low-RKLT is investigated using multi-lifting in this paper. Combined with ROI coding method, we have proposed a progressive lossy-to-lossless ROI compression method for three dimensional (3D) medical images with high performance. In our proposed method Low-RKLT is used for the inter-frame decorrelation after SA-DWT in the spatial domain. Simulation results show that, the proposed method performs much better in both lossless and lossy compression than 3D-DWT-based method.

This paper proposes an image coding method based on adaptive downsampling which not only uses the pixel redundancy but also considers visual redundancy. At the encoder side, codec adaptively chooses some smooth regions of the original image to downsample, and then overlapped transform with selectivity, block DCT and adaptive-shape DCT (SA-DCT) are used against the image after being downsampled. For the incomplete transformed image, OB-SPECK is adopted to code. At the decoder side, in order to reduce the computational complexity, we use the simple cubic interpolation which not only is very suitable to the downsampled regions but also enhances greatly the real time of this coding system. Experimental results shows the proposed method outperforms JPEG2000, SPECK, SPIHT, and LT+SPECK at low bit rates.

We proposed a new transform scheme of multiplierless reversible time-domain lapped transform and Karhunen-Loeve transform (RTDLT/KLT) for lossy-to-lossless hyperspectral image compression. Instead of applying discrete wavelet transform (DWT) in the spatial domain, RTDLT is applied for decorrelation. RTDLT can be achieved by existing discrete cosine transform and pre- and postfilters, while the reversible transform is guaranteed by a matrix factorization method. In the spectral direction, reversible integer low-complexity KLT is used for decorrelation. Owing to completely reversible transform, the proposed method can realize progressive lossy-to-lossless compression from a single embedded code-stream file. Numerical experiments on benchmark images show that the proposed transform scheme performs better than 5/3DWT-based methods in both lossy and lossless compressions, comparable with the optimal 9/7DWT-FloatKLT-based lossy compression method.

A progressive image compression scheme is investigated using reversible integer discrete cosine transform (RDCT) which is derived from the matrix factorization theory. Previous techniques based on DCT suffer from bad performance in lossy image compression compared with wavelet image codec. And lossless compression methods such as IntDCT, I2I-DCT and so on could not compare with JPEG-LS or integer discrete wavelet transform (DWT) based codec. In this paper, lossy to lossless image compression can be implemented by our proposed scheme which consists of RDCT, coefficients reorganization, bit plane encoding, and reversible integer pre- and post-filters. Simulation results show that our method is competitive against JPEG-LS and JPEG2000 in lossless compression. Moreover, our method outperforms JPEG2000 (reversible 5/3 filter) for lossy compression, and the performance is even comparable with JPEG2000 which adopted irreversible 9/7 floating-point filter (9/7F filter).