Facial expression recognition (FER) remains challenging under pose, illumination, and occlusion. This work presents CCFER, a dual‐stream framework that couples explicit edge maps with appearance features. Grayscale faces undergo morphological closing followed by opening (5 × 5), then Canny with locally adaptive thresholds to produce clean edges for an edge branch; both streams use Dual‐Direction Attention Mixed Feature Networks (DDAMFN). Multilevel fusion employs adaptively spatial feature fusion (ASFF), followed by Efficient Local Attention (ELA) and multihead attention (MHAtt) before classification. CCFER attains 92.19% on RAF‐DB, 91.24% on FERPlus, and 67.32% on AffectNet‐7, matching or approaching the recent state of the art with balanced cross‐dataset performance. Controlled ablations (parameter‐matched single‐stream, random‐noise edges) confirm gains stem from semantic contours, and efficiency measurements show modest overhead in parameters, GFLOPs, and latency, supporting practical deployment.

ABSTRACT In this paper, we propose a machine learning‐driven model to optimize panoramic video transmission for multiple users in virtual reality environments. The model predicts users' future field of view (FOV) using historical head orientation data and video saliency information, enabling targeted video delivery based on individual perspectives. By segmenting panoramic videos into tiles and applying a pyramid coding scheme, we adaptively transmit high‐quality content within users' FOVs while utilizing lower‐quality transmissions for peripheral regions. This approach effectively reduces bandwidth consumption while maintaining a high‐quality viewing experience. Our experimental results demonstrate that combining user viewpoint data with video saliency features significantly improves long‐term FOV prediction accuracy, leading to a more efficient and user‐centric transmission model. The proposed method holds great potential for enhancing the immersive experience of panoramic video streaming in VR, particularly in bandwidth‐constrained environments.

Microservices architecture has garnered extensive attention for its stability and scalability. However, in the complex and dynamic landscape of microservices systems, a incident in one service can propagate to others, resulting in significant economic losses and degraded user experiences. Therefore, the effective and precise localization of incidents in microservices systems becomes a critical concern. Previous research has leveraged runtime data (logs, metrics, call traces) and historical incident data to assist in root cause localization. However, due to the scarcity of business incidents (those causing severe impacts on business operations) and the fact that many incidents are reported by users, relevant run-time data and sufficient historical data are often unavailable, rendering previous methods impractical. In response to this challenge, we propose an online contrastive learning-based method for root cause localization of business incidents(OCRCL). We fully exploit incident tickets and the static dependency graph of services, integrating both textual semantic information and structural information from the dependency graph to discover root causes. Furthermore, we suggest that online contrastive learning can exhibit excellent performance with limited data and enable real-time model updates, making it better suited for industrial scenarios. Our approach demonstrates significant improvements over baseline methods across three real-world industrial datasets, highlighting its effectiveness in root cause localization.

In online service systems, a majority of incidents are caused by changes, which can influence user experience and cause huge economic loss. Experiences with a real-world, large-scale online service system show that more than half of the change-induced incidents are reported by users. Identifying root-cause changes for these incidents is challenging due to the inherent gap between user-perceived functional-level incident information and component-level change details. Inadequate causal knowledge also brings challenges. In this paper, we propose a novel causal knowledge mining based approach aiming at root-cause change identification for user-reported incidents named Raccoon. To bridge the gap between incidents and changes, it utilizes the fault tree and software product line to represent incidents and changes at the user-perceived functional level. They are also used as the backbone of causal knowledge. To overcome the lack of causal knowledge, Raccoon adopts efficient knowledge extraction and inference methods. Moreover, Raccoon provides recommendations at the software product line and change granularity to meet diverse demands of incident triage and root-cause change identification scenarios in incident management. We evaluate Raccoon on a real-world dataset collected in a large-scale online service system. The result shows that Raccoon significantly outperforms the state-of-the-art baseline approaches, which proves its effectiveness.

In online service systems, software changes cause a majority of incidents (i.e., unplanned interruptions and outages). Managing change-induced incidents efficiently is crucial for ensuring the reliability and availability of online service systems. Understanding the incidents can help improve change-induced incident management. The task is challenging because the life cycle of change-induced incidents is complicated due to diverse change deployment and incident resolution procedures. Detailed records of the incidents and changes, together with a comprehensive analysis, are needed to gain an in-depth understanding. In this paper, we conduct a qualitative and quantitative study on 231 change-induced incidents in a real-world, large-scale online service system. Detailed change tickets and incident timeline in the post-mortems provides extensive information about the incident life cycle, enabling us to understand each incident in depth. Based on the data, we give a generic model of the complicated life cycle of change-induced incidents. Following the model, we systematically study the whole life cycle of the incident, including the introduction and resolution stages, and answer what affects the efficiency of resolution. We obtain 9 major findings from our study. Based on the findings, we discuss existing techniques and promising future directions for improving change-induced incident management.

The blur kernel estimated by a blind deblurring algorithm is hardly to be error-free. The blur kernel error is usually ignored in the nonblind deconvolution stage and may result in severe artifacts or other negative effects. In addition, the bias hidden in the blurry image formation model has not been found and investigated due to ignoring the existence of the blur kernel errors. To this end, we develop a nonblind deconvolution method by bias correction for inaccurate blur kernels in this article, which are constructed on the basis of the classic errors-in-variables (EIVs) model. First we analyze in detail the bias caused by errors of inaccurate blur kernel from blurry image formation model. Next, the latent sparsity property of jointed latent image and errors of inaccurate blur kernel is counted statistically, which is imposed as a new regularization term. Then, the objective function of the new nonblind method is established, and an alternative minimization algorithm is derived and employed to estimate the latent clear image. Furthermore, a filtering method is utilized to modify the bias term, which is added to amend intermediate image in each iteration. Finally, extensive experiments with benchmark datasets and blurred micro-nano satellite remote sensing images are carried out to evaluate the proposed method. Experimental results demonstrate that the proposed method can obtain high-quality restored images, and it is comparable to or even better than some state-of-the-art nonblind deconvolution methods.

Image matching for blurred images is one of the most important and frequently discussed topics. In order to improve the precision of image matching for blurred images, we add a deblurring process before matching. Due to the ill-posedness of the deblurring problem, we use an image sparsity prior combined with patch-wise minimal and maximal pixel of latent image. Half quadratics splitting algorithm is applied under the maximum a posterior (MAP) framework. Six classical feature descriptors are utilized to implement the image matching. Five evaluation indexes are used to test the effect of the proposed deblurring algorithm on motion blurred remote sensing images matching. Experimental results show that the proposed deburring method can impair the influence of motion blur and improve the precision of matching.

The spatial modulation coincidence imaging (SMCI), as a novel kind of microwave coincidence imaging method, is proposed in this paper. The SMCI system provides a new way to produce the time-space independent signal instead of multitransmitter architecture with wideband randomly modulated signal in radar coincidence imaging. Due to some special features, metamaterial plate is utilized as the reflective surface to modulate the incident signal to construct random radiation field. The resolution of SMCI system is analyzed under large viewing angle with two different transmitting signals. Reflective surface is nonuniformly divided to derive the expression of resolution. The analysis results show that the resolution of SMCI system is mainly determined by the size of reflective surface and center frequency, which is similar to the traditional aperture. The SMCI system is low cost and flexible in design. Simultaneously, it can avoid the synchronization problem between subsources. Moreover, the SMCI system can achieve the resolution of space target through single transmitter-single receiver radar system. High-resolution image can be reconstructed since the tests are nonlinear. Finally, a series of simulation experiments is presented based on the nondirect-viewing scene we proposed. Using the algorithm based on a compressed sensing theory, we reconstructed the target image with high resolution.

High-resolution satellite images (HRSIs) obtained from linear array charge-coupled device sensors always suffer from geometric instability in the presence of attitude jitter. Therefore, detection and compensation of spacecraft attitude jitter in both the cross-track and along-track directions are crucial to improve the geometric accuracy of HRSIs. A number of reports have been made on the detection and estimation of cross-track attitude jitter. However, the detection of the attitude jitter in the along-track direction is more complicated due to the impact of topographic change. This paper presents a novel approach to achieve accurate estimation of the along-track attitude jitter by eliminating the influence of topographic information based on the back-projection residuals of three-line-array (TLA) images. The principle of detection and estimation of along-track attitude jitter is described, and the proposed approach consists of three main components as follows: 1) dense image matching of the TLA images using a comprehensive matching strategy; 2) detection of the back-projection residuals in the line direction caused by attitude jitter; and 3) estimation of the along-track attitude jitter from the back-projection residuals using a genetic algorithm. Experiments were conducted using China's Ziyuan-3 (ZY-3) TLA images, and the experimental results reveal that the frequency of the attitude jitter in the along-track direction ranges between 0.6 and 0.7 Hz, which is consistent with the frequency in the cross-track direction observed in our previous study. In addition, a comparison of the results of the proposed approach with those from direct attitude observations shows good consistency, with as little as 0.1-pixel disparity, which demonstrates the feasibility and reliability of the proposed approach. Furthermore, the geometric accuracy is further improved from a pixel level to a subpixel level and the periodic trend is removed with the compensation of the estimated attitude jitter in addition to the conventional affine compensation, which validates the potential of the proposed approach for geometric accuracy improvement with ZY-3 TLA images.

With the rapid development of geospatial data capture technologies such as the Global Positioning System, more and higher accuracy data are now readily available to upgrade existing spatial datasets having lower accuracy using positional accuracy improvement (PAI) methods. Such methods may not achieve survey-accurate spatial datasets but can contribute to significant improvements in positional accuracy in a cost-effective manner. This article addresses a comparative study on PAI methods with applications to improve the spatial accuracy of the digital cadastral for Shanghai. Four critical issues are investigated: (1) the choice of improvement model in PAI adjustment; five PAI models are presented, namely the translation, scale and translation, similarity, affine, and second-order polynomial models; (2) the choice of estimation method in PAI adjustment; three estimation methods in PAI adjustment are proposed, namely the classical least squares (LS) adjustment, which assumes that only the observation vector contains error, the general least squares (GLS) adjustment, which regards both the ground and map coordinates of control points as observations with errors, and the total least squares (TLS) adjustment, which takes the errors in both the observation vector and the design matrix into account; (3) the impact of the configuration of ground control points (GCPs) on the result of PAI adjustment; 12 scenarios of GCP configurations are tested, including different numbers and distributions of GCPs; and (4) the deformation of geometric shape by the above-mentioned transformation models is presented in terms of area and perimeter. The empirical experiment results for six test blocks in Shanghai demonstrated the following. (1) The translation model hardly improves the positional accuracy because it accounts only for the shift error within digital datasets. The other four models (i.e., the scale and translation, similarity, affine, and second-order polynomial models) significantly improve the positional accuracy, which is assessed at checkpoints (CKPs) by calculating the difference between the updated coordinates transformed from the map coordinates and the surveyed coordinates. On the basis of the refined Akaike information criterion, the two best optimal transformation models for PAI are determined as the scale and translation and affine transformation models. (2) The weighted sum of square errors obtained using the GLS and TLS methods are much less than those obtained using the classical least squares method. The result indicates that both the GLS and TLS estimation methods can achieve greater reliability and accuracy in PAI adjustment. (3) The configuration of GCPs has a considerable effect on the result of PAI adjustment. Thus, an optimal configuration scheme of GCPs is determined to obtain the highest positional accuracy in the study area. (4) Compared with the deformations of geometric shapes caused by the transformation models, the scale and translation model is found to be the best model for the study area.

The highly skewed sex ratio at birth (SRB) in China has stimulated numerous studies. However, the geographic distribution of SRB is seldom investigated, particularly at the county level. The need for an understanding at this level has increased since the Chinese government initiated its ‘Care for Girls’ campaign to improve the survival rate of females. This campaign has been initiated in a set of pilot counties. In this article we assess the effectiveness of the set of pilot counties in Shandong province and propose two alternate configurations. To do this, we first assess the spatial distribution of the SRB values by county in Shandong, expressed as a z-score (zSRB) after correcting for the biologically expected SRB value and population size of zero-aged children. A local Moran's Ii analysis of the zSRB values indicates a significant high–high cluster in the southwest of the province. The Ii , zSRB and female deficit (the difference of the observed from biologically expected number of zero-aged females) were then used to define two alternate configurations for the pilot counties. A comparison of the current and alternate configurations against a Monte Carlo randomisation analysis shows that the current configuration is significantly different from a random selection (p < 0.05) for the two criteria of maximising the aggregate female deficit and maximising the zSRB. Although this is a good result, both alternate configurations were more significant (p < 0.001), and therefore represent potentially better configurations for the campaign given the criteria used. The spatial analysis approach developed here could be used to improve the effectiveness of the Care-for-Girls campaign in Shandong province, and elsewhere in China.