Intracerebral hemorrhage (ICH) often leads to high disability and mortality rates, and accurate and rapid segmentation of hematoma regions is crucial for condition assessment and treatment planning. Existing neural network methods are unsatisfactory due to the irregularity of hematoma morphology and location, and limited generalization performance. To address these issues, this paper proposes an improved U-Net architecture, which incorporates two innovative modules: an Adaptive High-Frequency Domain Enhancement module and a Context-Aware Similarity Fusion module. The former leverages ICH's highsignal characteristics through frequency-domain enhancement and intensity-guided weighting to selectively enhance ICH regions, thereby improving the network's sensitivity to ICH. The latter enables efficient integration of multi-level contextual information by calculating feature similarity between decoder levels and adaptively filtering and fusing features according to spatial consistency, thereby improving segmentation performance. Experimental results demonstrate that our method achieves improved performance across multiple evaluation metrics compared to existing methods, and provides a more reliable tool for automated diagnosis of ICH.

Surgical phase recognition is a fundamental component of computer-assisted surgery, providing essential contextual understanding of complex surgical workflows. This paper introduces herein SSTANet, an end-to-end network designed for surgical phase recognition and anchored in a novel Surgical Spatio-Temporal Aggregation (SSTA) module. Unlike conventional 3D CNNs or Transformer-based temporal aggregators that rely on heavy convolutional kernels or global self-attention, SSTA adopts a parallel temporal modeling and lightweight fusion strategy to jointly capture local and long-term temporal dependencies while preserving spatial cues. Specifically, SSTA integrates multi-level representations from a ResNet50-based spatial feature extractor and a hybrid Bi-LSTM + TCN temporal feature extractor, followed by an adaptive spatio-temporal projection layer that efficiently balances model performance and computational cost. This design enables fine-grained phase understanding without the complexity of 3D convolutions or highdimensional attention mechanisms. Comprehensive experiments on the Cholec80 dataset demonstrate the effectiveness of the proposed approach, achieving 88.88 % accuracy, 74.07 % Jaccard, 84.43 % precision, and 85.84 % recall, outperforming several state-of-the-art methods. The results confirm that the proposed SSTA module effectively bridges spatial-temporal information and enhances recognition robustness under complex surgical conditions, offering practical potential for real-time computerassisted surgery systems.

Binary (0-1) integer programming (BIP) is pivotal in scientific domains requiring discrete decision-making. As the advance of AI computing, recent works explore neural network-based solver for integer linear programming (ILP) problems. Yet, they lack scalability for tackling nonlinear challenges. To handle nonlinearities, state-of-the-art Branch-and-Cut solvers employ linear relaxations, leading to exponential growth in auxiliary variables and severe computation limitations. To overcome these limitations, we propose BIPNN, an unsupervised learning framework to solve BIP problems via hypergraph neural networks (HyperGNN). Specifically, (i) BIPNN reformulates BIPs-constrained, discrete, and nonlinear (sin, log, exp) optimization problems-into unconstrained, differentiable, and polynomial loss functions. The reformulation stems from the observation of a precise one-to-one mapping between polynomial BIP objectives and hypergraph structures, enabling the unsupervised training of HyperGNN to optimize BIP problems in an end-to-end manner. On this basis, (ii) we propose a GPU-accelerated and continuous-annealing-enhanced training pipeline for BIPNN. The pipeline enables BIPNN to optimize large-scale nonlinear terms in BIPs fully in parallel via straightforward gradient descent, thus significantly reducing the training cost while ensuring the generation of discrete, high-quality solutions. Extensive experiments on synthetic and real-world datasets highlight the superiority of our approach.

Satellite remote sensing cooperation is essential for ensuring efficient data transmission in real-time applications. Network traffic prediction plays a crucial role in optimizing data transmission strategies, managing congestion, and reducing network latency. However, current research work on network traffic prediction frequently fails to fully exploit the complex spatial-temporal dependencies inherent in satellite network traffic. To address this limitation and improve the accuracy of long-term network traffic prediction, we propose an Efficient Spatiotemporal Prediction Transformer (ESPformer) for dynamic data transmission in cooperative satellite remote sensing. The proposed scheme not only considers propagation delays but also captures the temporal and spatial relationships among the network traffic. In particular, we design a spatial-temporal multi-head attention mechanism within an encoder-decoder transformer to capture the dynamic spatial dependencies and predict the network topology and its parameters, including traffic flow and bandwidth. By leveraging historical traffic data and the network traffic conditions, the model estimates expected queuing delays. Finally, based on the volume of traffic predicted and the changes in network conditions, we dynamically adjust the transmission strategies to maintain an efficient relaying mechanism. Therefore, our model enables an adaptive transmission strategy and offers an optimal delay reduction in real-time satellite data transmission. Extensive experiments conducted on four well-known traffic datasets demonstrate that the ESPformer significantly outperforms state-of-the-art baselines across all key performance metrics.

Hyperspectral image (HSI) classification with limited training samples remains a challenging task due to the high cost and time consumption of collecting labeled samples. In recent years, semi-supervised image classification methods have garnered widespread attention, as they address the scarcity of labeled samples and enhance classification accuracy by leveraging the inherent relationships between labeled and unlabeled samples. In this paper, we introduce a semi-supervised classification method based on multi-scale spectral-spatial graph attention network (MSSGAT). We construct multiple neighborhood graphs with different scales, where all samples (including training and testing samples) are represented as nodes, and exploit the spectral-spatial features with different receptive fields. These features are then fed into the graph attention network to learn attention coefficients between neighboring nodes and self-attention coefficients for target nodes, which are further aggregated to extract more discriminative features for enhancing HSI classification accuracy. The experimental results on the Indian Pines, Salinas and Pavia University datasets demonstrate its competitive performance. The overall accuracy ( OA ) scores are 99.36 %, 99.59 % and 96.85 %, respectively, which exceed the state-of-the-art ( SOTA ) models.

The booming crowdsourced delivery leverages networks of local, part time couriers to meet growing customer demands for increasingly speedy shipping. By conducting city-wide low-cost delivery networks, crowd-sourced enabled instant delivery platforms have fought to figure out a way to offer instant shipping at low price. Nevertheless, limited ubiquitous computing infrastructures impede the discovery of equilibrium between cost reduction and sufficient shipping capacity. In this article, we present our endeavor towards understanding and designing a dynamic pricing framework based on a large-scale food-delivery platform. To reduce delivery fees in capacity overload regions and resolve the decline of service quality caused by insufficient shipping capacity, we address the challenge of quantifying the inter-regional shipping capacity disparities by accordingly applying a variety of techniques, including cooperative game theory, probability analysis and address entity recognition. In particular, we provide an$O(n^{2})$algorithm for the cooperative game based cost allocation, of which the exponential computational complexity$O(n^{2}2^{n})$is generally considered a major challenge in real-world applications. City-wide experiments in Nanjing and Shenzhen illustrate that we achieve a 10% decrease in delivery fees under overload shipping capacity and a 3% increase in completion rates in customer locations with poor service quality.

Graph cuts is an image segmentation method by which the region and boundary information of objects can be revolved comprehensively. Because of the complex spatial characteristics of high-dimensional images, time complexity and segmentation accuracy of graph cuts methods for high-dimensional images need to be improved. This paper proposes a new three-dimensional multilevel banded graph cuts model to increase its accuracy and reduce its complexity. Firstly, three-dimensional image is viewed as a high-dimensional space to construct three-dimensional network graphs. A pyramid image sequence is created by Gaussian pyramid downsampling procedure. Then, a new energy function is built according to the spatial characteristics of the three-dimensional image, in which the adjacent points are expressed by using a 26-connected system. At last, the banded graph is constructed on a narrow band around the object/background. The graph cuts method is performed on the banded graph layer by layer to obtain the object region sequentially. In order to verify the proposed method, we have performed an experiment on a set of three-dimensional colon CT images, and compared the results with local region active contour and Chan–Vese model. The experimental results demonstrate that the proposed method can segment colon tissues from three-dimensional abdominal CT images accurately. The segmentation accuracy can be increased to 95.1% and the time complexity is reduced by about 30% of the other two methods.