Recently, color point cloud registration has begun to receive attention. Unlike geometric-only point clouds, color point clouds incorporate additional color information. Therefore, color point cloud registration can achieve higher accuracy than geometric-only point cloud registration. Despite the success, existing methods are computationally intensive due to the high resource demands of Transformer. In this paper, we propose a Mamba architecture based registration algorithm, CGFMamba-PCR, with color-geometric fusion. Specifically, we propose CGFMamba, a novel color-geometric fusion Mamba network for color point cloud registration, which enhances feature representation through color-geometric guided point ordering and positional encoding. For the input color point cloud pair, they are passed through CGFMamba based feature extraction module to obtain their corresponding features. Then, these features are passed through feature matching and outlier rejection modules to obtain final registration result. Furthermore, an ordering method for the Mamba architecture is proposed that clusters color hue and sorts spatial coordinates. Experiments on Color3DMatch and Color3DLoMatch datasets demonstrate that the proposed algorithm outperforms the state-of-the-art (SOTA) methods. The code of the proposed algorithm will be open-sourced upon acceptance of this paper.

Electrocardiography (ECG) is a fundamental tool for cardiovascular disease detection, yet accurately identifying QRS complexes in ultra-short recordings remains challenging. Existing methods often rely on R-peak detection or fixed-length segmentation, leading to imprecise onset and offset localization and requiring extended recordings for reliable performance. To address these challenges, we propose MAC-Net, a Multi-Scale Atrous Convolutional Network for precise and real-time QRS detection in 1-second single-lead ECGs. To validate its effectiveness, we curated a high-quality clinical dataset, addressing the scarcity of ultra-short ECG resources. By leveraging hierarchical multi-scale feature extraction and atrous convolutions, MAC-Net achieves state-of-the-art performance with a sensitivity of 0.970 and a positive predictive value of 0.995. Additionally, it generalizes effectively for R-peak detection across four public datasets under zero-shot learning conditions, demonstrating robust adaptability and clinical applicability.

Height estimation from single aerial images plays a crucial role in various remote sensing applications. However, due to the high capturing altitude, aerial images often lack detailed textures and primarily contain building roofs, which limits the generalization of models trained on one dataset to another. This paper investigates methods to enhance cross-dataset generalization for height estimation models. Our contributions are two-fold: First, we introduce a pretrain-finetune paradigm that enables the transfer of a height estimation model across datasets, requiring only a small number of images from a new dataset. Second, we propose two fine-tuning strategies: a supervised method for generalization when limited height ground-truth data are available, and a self-supervised approach that utilizes image pairs and their corresponding RPC models to fine-tune the model in the absence of ground-truth height data. We validate the effectiveness of our methods through extensive experiments on four public datasets—Vaihingen, Potsdam, DFC2019, and MVS3DM—demonstrating their ability to generalize height estimation models across diverse datasets.

Sketch-based image retrieval (SBIR) plays a crucial role in various visual search and recognition applications, yet the sparse and abstract nature of sketches poses significant challenges to accurate retrieval. Existing methods often struggle to effectively capture both global contextual information and fine-grained details simultaneously, while relying on single-metric learning, resulting in suboptimal performance. To address these issues, we propose NFF-MML-Net, a novel method that integrates non-local feature fusion and mixed multi-metric learning for enhanced sketch-based image retrieval. The non-local feature fusion improves feature extraction by incorporating non-local dependencies across channels and spaces via a non-local attention mechanism, while also performing weighted fusion of non-local features and handcrafted features, significantly enhancing feature representation and uniqueness. In the mixed multi-metric learning framework, a hybrid triplet loss function is designed, which combines Euclidean distance, cosine similarity, and a directional regularization term to capture both distance and directional differences, guiding the model towards more precise updates. Extensive experiments on the QMUL-ChairV2 and QMUL-ShoeV2 public datasets demonstrate the retrieval performance of NFF-MML-Net, achieving Recall@1 scores of 74.73% and 41.37% on the respective datasets.

Immersive communications, such as the Extended Reality (XR) and Cloud Gaming (CG), is one of six usage scenarios in the 6th generation (6G) [1]. These applications demand stringent requirements on the frame level delay and frame level reliability. At the application layer, techniques such as Forward Error Correction (FEC) are employed to ensure successful frame decoding even if some encoded packets are not received in time. However, the technical benefits of application layer FEC cannot actually be realized as although the upper layer (i.e., application layer) can tolerate errors, lower layer (i.e., physical (PHY) layer) of wireless network strives for 100% reliability. To address this contradiction, this paper proposed a Medium Access Control (MAC) sub Protocol Data Unit (subPDU) self-decoding method, where receiver (RX) can independently decode MAC subPDU(s) corresponding to a Code Block (CB) in a Transport Block (TB), without relying on the successful reception of preceding CBs. Simulation results demonstrate that the proposed method can increase the number of users whose frame decoding success rate exceeds a certain threshold in a cell (i.e., cell capacity) by 20%, the overall user experience for immersive services can be greatly improved.

To minimize the disturbance of the Tunnel Boring Machine (TBM) cutterhead on the surrounding rock during the coal mine roadway excavation process and ensure that the cutterhead rotation speed achieves fast tracking performance with minimal overshoot, we propose a direct adaptive robust control method for the cutterhead rotation speed hydraulic system based on inversion design. This method considers the strong disturbances such as loads and motion affecting the cutterhead hydraulic drive system, as well as the uncertainties in the cutterhead model. We establish the nonlinear model of the cutterhead hydraulic drive system and employ virtual control to reduce the model order. Using Lyapunov functions, we ensure the stability of the entire system and derive the control law for the cutterhead rotation speed controller, along with parameter-adaptive laws acting as parameter estimators. We validate the effectiveness of the proposed control strategy through joint simulation using AMESim and Simulink. The results show that the designed cutterhead rotation speed controller achieves high tracking accuracy and good adaptability.

3D modeling of the Earth's surface is an important topic and finds its various applications in the remote sensing communities. Due to the imaging complexity of optical remote sensing images, the process of creating a digital surface model of the Earth from multi-view optical remote sensing images is both time-consuming and challenging, especially when dealing with large areas. In this work, we propose a deep learning-based approach, called RS-MVSNet, for inferring a digital surface model from multi-view optical remote sensing images. In order to extend state-of-the-art learning-based multi-view stereo techniques to optical remote sensing images, a differentiable affine warping is designed for the first time, which utilizes an affine to Euclidean upgraded camera model to model the mapping relationship between small-sized remote sensing image tiles and their corresponding 3D local scenes in Euclidean space. Based on the differentiable affine warping, RS-MVSNet is abstracted from the complexities associated with remote sensing imaging and inherits generic components for deep learning based multi-view stereo, including multi-scale deep feature extraction, pyramid cost volume construction, regularization and regression. Moreover, an affine epipolar guided feature aggregation module is constructed in the proposed RS-MVSNet framework to accurately aggregate high-resolution remote sensing image features along affine epipolar lines into a finite cost volume. Extensive experiments are conducted on two public datasets, namely MVS3DM and US3D datasets, and our proposed RS-MVSNet shows promising results in terms of accuracy and efficiency.

Erlang is a concurrent functional programming language and is used to program concurrent, distributed and fault-tolerant systems. In this paper, we study the higher-order communication mechanism of Erlang's functional programming, which is actually achieved by means of sending the references to functions in the current implementation. We propose an improved design. In this design, we realize purely higher-order communication, i.e., strict program-passing. Moreover, we realize automatic determination of module dependency and simplification of communicating anonymous function, i.e., only sending the definition of anonymous function instead of the entire module. Finally, experiments are designed to test the implementation.