In remote sensing image segmentation, recognizing buildings is challenging when the visual evidence from pixels is weak or when buildings belong to small, spatially structured objects. To address this issue, we propose a structure-prior guided adaptive context selection network (SGACS-Net) for remote sensing semantic segmentation. The core is to use structure-prior knowledge to dynamically capture prior contextual information and higher-order object structural features, thereby improving the accuracy of remote sensing building segmentation. First, an adaptive context selection module is designed. By dynamically adjusting the spatial sensing field, this module effectively models the global long-range context information dependencies. It captures varying context information of buildings at different scales, thereby enhancing the network’s ability to extract building feature representations. Second, a structure-prior guided variable loss function is proposed. It utilizes the structural features of building points, lines, and surface to identify key regions. By leveraging advanced structure-prior knowledge, it enhances the network’s ability to express structural features. Experimental results on two datasets show that the proposed SGACS-Net outperforms other typical and state-of-the-art methods in terms of remote sensing semantic segmentation performance.

Attention-based Transformer model has achieved remarkable performance in various artificial intelligence fields, while the attention computation, being a combination of matrix multiplication and softmax function, is still sub-optimized in terms of hardware implementation. Normally, it needs 3 pass of input memory access to compute attention, and the on-chip storage requirement is coupled with the input length, both of which pose significant memory issues. Further, the computation burden is heavy for long inputs. This paper proposes an algorithm-hardware co-design for attention. On algorithm side, it uses a linear-softmax fused kernel to fuse the matrix multiplication and non-linear functions, which enables high on-chip memory source utilization. On hardware side, it shows an accelerator named BLADE with identical partial product removing, which eliminates unnecessary computation by exploiting math feature of softmax. Experiment on ViT, Swin Transformer, GPT-2 and LLaMA2 show that the proposed design achieves 10.6% - 18.7% of energy efficiency improvement compared to state-of-the-art Flash Attention implementations.

The time step functions as a crucial temporal unit for simulating neuronal dynamics within spiking neural networks, which play a significant role in neuromorphic computing systems. Efficient management of these time steps is vital to ensure model accuracy while optimizing overall system performance. As system scale increases, variations in hardware across subsystems and their asynchronous operations create challenges in achieving effective time step control. To address this issue, this paper proposes an innovative framework for managing time steps in large-scale neuromorphic systems. This framework allows subsystems to dynamically adjust their time step lengths according to computational loads and to perform look-ahead computations. Such a strategy effectively reduces the overhead related to time step synchronization, enhancing system efficiency. Additionally, the paper introduces a safeguard mechanism to ensure the system’s reliability. Experimental results indicate that the proposed framework sustains the correct long-term operation of the system and improves model execution performance by 8.88% to 27.15% when compared to existing methods.

A low-cost high-responsivity design of Si/SiGe Heterojunction Bipolar Phototransistor (HPT) built in an industrial 55nm-BiCMOS technology is integrated in a cascode configuration with a bipolar bipolar transistor (HBT) to be used for microwave-photonic communication receivers. Performances of both the cascode pair and the single Heterojunction Bipolar Phototransistor are compared. At high-frequency, the cascode circuit has better performances than the single HPT stage. At 850-nm wavelength, the pair can reach a low-frequency responsivity a low-frequency responsivity of 3.24 A/W and a bandwidth of 0.69 GHz when coupling losses are corrected.

not-yet-known not-yet-known not-yet-known unknown Arthroscopy is a minimally invasive surgical procedure used to diagnose and treat joint problems. The clinical workflow of arthroscopy typically involves inserting an arthroscope into the joint through a small incision, during which surgeons navigate and operate largely by relying on their visual assessment through the arthroscope. However, the arthroscope’s restricted field of view and lack of depth perception pose challenges in navigating complex articular structures and achieving surgical precision during procedures. Aiming at enhancing intraoperative awareness, we present a robust pipeline that incorporates simultaneous localization and mapping, depth estimation, and 3D Gaussian splatting to realistically reconstruct intra-articular structures solely based on monocular arthroscope video. Extending 3D reconstruction to Augmented Reality (AR) applications, our solution offers AR assistance for articular notch measurement and annotation anchoring in a human-in-the-loop manner. Compared to traditional Structure-from-Motion and Neural Radiance Field-based methods, our pipeline achieves dense 3D reconstruction and competitive rendering fidelity with explicit 3D representation in 7 minutes on average. When evaluated on four phantom datasets, our method achieves RMSE = 2.21 mm reconstruction error, PSNR = 32.86 and SSIM = 0.89 on average. Because our pipeline enables AR reconstruction and guidance directly from monocular arthroscopy without any additional data and/or hardware, our solution may hold the potential for enhancing intraoperative awareness and facilitating surgical precision in arthroscopy. Our AR measurement tool achieves accuracy within 1.59 ± 1.81 mm and the AR annotation tool achieves a mIoU of 0.721.

Inverse Synthetic Aperture Radar (ISAR) is a crucial radar imaging technique that relies on the relative motion between the radar and the target to produce high-resolution images. However, traditional ISAR methods are highly sensitive to inaccuracies in estimating rotational parameters, such as roll, pitch, and yaw. Errors in these estimates can result in significant image degradation. In this paper, we present a novel approach for imaging dynamically rotating scenes using a multistatic ISAR setup, where we estimate motion parameters through a Differential Semblance Optimisation (DSO) criterion. This is done by minimising discrepancies between images formed from multiple transmitter-receiver pairs, our method delivers focused images with improved estimates of the yaw rotation parameter of the targets. We demonstrate the effectiveness of this approach through a series of numerical experiments, highlighting its robustness in both noise-free and noisy environments, and discuss its potential for enhancing ISAR imaging in complex scenarios.

Fluid antenna multiple access (FAMA) is a concept capable of massive connectivity on the same physical channel without the need of precoding or interference cancellation, by exploiting a super-high dimensional received signal in the spatial domain from fluid antenna system (FAS). This letter investigates the use of 5G New Radio (NR) Modulation Coding Scheme (MCS) for FAMA systems to improve its connectivity over block fading channels. In particular, we propose an iterative decoding receiver with iterative interference covariance estimation for the coded FAMA system. Extrinsic information transfer (EXIT) chart is used to analyze the performance of the proposed iterative receiver. Both EXIT chart analysis and numerical results show that the proposed receiver outperforms the one that uses fixed covariance in the block fading channel. Also, the results indicate that coded FAMA can serve 100 user terminals (UTs) at the rate of 0.5 bit/s/Hz under rich scattering.

In complex dynamic environments, robot path planning faces the challenge of multi-objective optimization, such as path length, smoothness and obstacle avoidance capability. To this end, this paper proposes an Improved Sparrow Search Algorithm (ISSA) based on chaotic initialization and golden positive cosine strategy for multi-objective path planning. Diverse initial populations are generated through chaotic mapping to enhance the global search capability and avoid falling into local optimum. The golden positive cosine strategy then optimizes individual position updates to accelerate convergence and ensure path smoothness. The results show that the proposed ISSA for path planning exhibits significant advantages in terms of path quality and convergence speed. After achieving global path optimization, for the dynamic obstacle environment, this paper adopts the Improved Dynamic Window Approach (IDWA) to achieve real-time obstacle avoidance, which dynamically adjusts the window size according to the robot speed, obstacle density, and target distance, and adaptively expands or shrinks the search space to improve the flexibility and efficiency of obstacle avoidance. Simulation results show that the method outperforms other algorithms in terms of path length, smoothness, obstacle avoidance ability and computational efficiency in both static and dynamic environments.

Despite the benefits of minimally invasive surgery, interventions such as laparoscopic liver surgery present unique challenges, like the significant anatomical differences between preoperative images and intraoperative scenes due to pneumoperitoneum, patient pose, and organ manipulation by surgical instruments. To address these challenges, we propose a method for intraoperative 3D reconstruction of the surgical scene, including vessels and tumors, without altering the surgical workflow. The technique combines Neural Radiance Field (NeRF) reconstructions from tracked laparoscopic videos with ultrasound 3D compounding. We evaluate the accuracy of our reconstructions on a clinical laparoscopic liver ablation dataset, consisting of laparoscope and patient reference poses from optical tracking, laparoscopic and ultrasound videos, as well as preoperative and intraoperative CTs. We propose a solution to compensate for liver deformations due to pressure applied during ultrasound acquisitions, improving the overall accuracy of the 3D reconstructions compared to the ground truth intraoperative CT with pneumoperitoneum. We train a unified NeRF from the ultrasound and laparoscope data, which allows real-time view synthesis providing surgeons with comprehensive intraoperative visual information for laparoscopic liver surgery.

not-yet-known not-yet-known not-yet-known unknown Deep learning models are used to minimize the number of polyps that goes unnoticed by the experts and to accurately segment the detected polyps during interventions. Although state-of-the-art models are proposed, it remains a challenge to define representations that are able to generalize well and that mediate between capturing low-level features and higher-level semantic details without being redundant. Another challenge with these models is that they are computation and memory intensive, which can pose a problem with real-time applications. To address these problems, we propose PlutoNet for polyp segmentation which requires only 9 FLOPs and 2,626,537 parameters, less than 10% of the parameters required by its counterparts. With PlutoNet, we propose a novel decoder consistency training approach that consists of a shared encoder, the modified partial decoder, which is a combination of the partial decoder and full-scale connections that capture salient features at different scales without redundancy, and the auxiliary decoder which focuses on higher-level semantic features. We train the modified partial decoder and the auxiliary decoder with a combined loss to enforce consistency, which helps strengthen learned representations. We perform ablation studies and experiments which show that PlutoNet performs significantly better than the state-of-the-art models, particularly on unseen datasets.

Background: Percutaneous nephrostomy can be an effective means of preventing irreparable renal damage from obstructive renal disease thereby providing patients with more time to access treatment to remove the source of the blockage. In sub-Saharan Africa, where there is limited access to treatments such as dialysis and transplantation, a nephrostomy can be life-saving. Training this procedure in simulation can allow trainees to develop their technical skills without risking patient safety, but still requires an ex-pert observer to provide performative feedback. In this study, we evaluate the feasibility of using video as an accessible method to assess skill in simulated percutaneous nephrostomy. Methods: Six novice urology residents and six expert urologists from Ouakam Military Hospital in Dakar, Senegal performed 4 nephrostomies each using our setup. We computed motion-based metrics for each trial from the predicted bounding boxes of a trained object detection network and compared these metrics between novices and experts. Results: We were able to measure significant differences in both ultrasound and needle handling between novice and expert participants. Additionally, we could measure performance changes within each group over multiple trials. Conclusions: Video-based skill assessment is a feasible and accessible option for providing trainees with quantitative feedback in sub-Saharan Africa.

The Segment Anything Model (SAM) is a powerful vision foundation model that is revolutionizing the traditional paradigm of segmentation. Despite this, reliance on prompting each frame and large computational cost limit its usage in robotically assisted surgery. Applications, such as augmented reality guidance, require little user intervention along with efficient inference to be usable clinically. In this study, we address these limitations by adopting lightweight SAM variants to meet the efficiency requirement and employing fine-tuning techniques to enhance their generalization in surgical scenes. Recent advancements in Tracking Any Point (TAP) have shown promising results in both accuracy and efficiency, particularly when points are occluded or leave the field of view. Inspired by this progress, we present a novel framework that combines an online point tracker with a lightweight SAM model that is fine-tuned for surgical instrument segmentation. Sparse points within the region of interest are tracked and used to prompt SAM throughout the video sequence, providing temporal consistency. The quantitative results surpass the state-of-the-art semi-supervised video segmentation method XMem on EndoVis2015 with 84.8 IoU and 91.0 Dice. In terms of efficiency, we tested our method on a single 4060/4090 GPU, achieving an over 25/90 FPS. Code is available at: https://github.com/wuzijian1997/SIS-PT-SAM

In percutaneous pelvic trauma surgery, accurate placement of Kirschner wires(K-wires) is crucial to ensure effective fracture fixation and avoid complications due to breaching the cortical bone along an unsuitable trajectory. Surgical navigation via mixed reality(MR) can help achieve precise wire placement in a low-profile form factor. Current approaches in this domain are as yet unsuitable for real-world deployment because they fall short of guaranteeing accurate visual feedback due to uncontrolled bending of the wire. To ensure accurate feedback, we introduce StraightTrack, an MR navigation system designed for percutaneous wire placement in complex anatomy. StraightTrack features a marker body equipped with a rigid access cannula that mitigates wire bending due to interactions with soft tissue and a covered bony surface. Integrated with an Optical See-Through Head-Mounted Display (OST HMD capable of tracking the cannula body, StraightTrack offers real-time 3D visualization and guidance without external trackers. In phantom experiments with two experienced orthopedic surgeons, StraightTrack improves wire placement accuracy, achieving the ideal trajectory within 5.26 ± 2.29 mm and 2.88 ±1.49°, compared to over 12.08 mm and 4.07° for comparable methods. As MR navigation systems continue to mature, StraightTrack realizes their potential for internal fracture fixation and other percutaneous orthopedic procedures.

Dynamic filtering is crucial in various applications, including imaging, communication, and biological detection. With advancements in micro-nano device fabrication, there is a growing demand for enhanced filtering performance. This paper presents a near-infrared single-wavelength tunable filter that utilizes polymer-dispersed liquid crystal combined with a Fabry-Perot resonant cavity. The device achieves a transmittance of 99% at 927 nm with a 24 nm full width at half maximum. Ultizing the electro - optical tunable properties of Dielectric layer, the transmission wavelength can be tuned from 913 to 969 nm by altering external voltage to it, resulting in a wavelength shift of 56 nm while maintaining high transmittance. This structure provides a reference for the dynamic control of light sensors and near-infrared dynamic imaging.

The estimation of the pose of surgical instruments is important in Robot-assisted Minimally Invasive Surgery (RMIS) to assist surgical navigation and enable autonomous robotic task execution. The performance of current instrument pose estimation methods deteriorates significantly in the presence of partial tool visibility, occlusions, and changes in the surgical scene. In this work, a vision-based framework is proposed for markerless estimation of the 6DoF pose of surgical instruments. To deal with partial instrument visibility, a keypoint object representation is used and stable and accurate instrument poses are computed using a PnP solver. To boost the learning process of the model under occlusion, a new mask-based data augmentation approach has been proposed. To validate our model, a dataset for instrument pose estimation with highly accurate ground truth data has been generated using different surgical robotic instruments. The proposed network can achieve submillimeter accuracy and our experimental results verify its generalisability to different shapes of occlusion.

Phased array antennas, with high directivity and low sidelobe levels, are extensively employed in radar and communication systems. As the array scale increases, the complexity and cost of controlling each element individually are increased significantly. To overcome this challenge and improve the efficiency, one of the effective engineering solutions is subarray partitioning. Traditional clustering algorithms, such as K-means, require a predefined number of subarrays, and limits the flexibility, especially in scenarios with complex geometries or irregular distributions. A subarray partitioning approach is presented in this paper using the Iterative Self-Organizing Data Analysis (ISODATA) algorithm, and the proposed scheme dynamically adjusts the number of subarrays based on the geometric characteristics of the array. According to numerical experiments results, good performance of sidelobe suppression, beamforming and partitioning flexibility have been achieved, and shows better characteristic compared with the K-means. This algorithm has potential application in large-scale phased arrays.

In this letter, a beam downtilt reconfigurable linear antenna array is presented for the demand of the electrical downtilt of 5G/6G macro base station antennas. The antenna is capable of fine-angle beam direction reconfiguration, allowing for improved adaptability in various environmental conditions. A hierarchical network design is employed, which incorporates essential elements such as phase shifter modules, power dividers, and couplers. These fundamental components are used to construct a five-element reconfigurable feeding network for the antenna array. The designed network ensures stable phase shifts and excellent impedance matching performance. By integrating the feeding network with the antenna elements, the system achieves electronic beam tilt adjustment across four states, enabling the linear array to tilt downward within a range of 6 - 12° around 3.5 GHz. This design provides enhanced beam control, making it suitable for next-generation macro base station applications in 5G and 6G networks.

To address the challenges posed by frequent source and load fluctuations in existing loop closing current calculation methods, this paper proposes an online loop closing current calculation method that considers source and load uncertainties. First, a dual-stack dynamic monitoring system is utilized to obtain real-time voltage and current variations before and after disturbances. Second, Thevenin's theorem is employed to build an equivalent model of the distribution network, simplifying the complex network into a combination of an independent voltage source and a series impedance. Then, the steady-state loop closing current is calculated based on the open-circuit voltage and equivalent impedance at both sides of the loop closing point. Next, the optimal frequency method is applied to determine the equivalent impedance and attenuation time constant at a specific frequency, achieving accurate calculation of the transient loop closing current. Finally, simulations are conducted to model the fluctuations in distributed generation and load, analyzing the steady-state and transient loop closing currents. The simulation results demonstrate that the proposed method accurately captures the effects of source and load fluctuations on the loop closing current in dynamic environments, with minimal calculation error, indicating its high practicality.