Tianmiao Wang

Concrete crack segmentation in bridge health monitoring is crucial for ensuring the safety and longevity of infrastructure. However, this task is complicated by challenges such as weak contrast, background interference, and multi-scale crack structures, which hinder traditional methods’ accuracy. This study introduces a novel Boundary-Sensitive Hybrid Attention Network (BSA-Net) designed to address these issues by combining a hierarchical Transformer encoder (Hiera-A), a multi-scale context module (Light-ASPP), and a boundary-aware decoder (BAD). The hierarchical encoder effectively captures multi-scale features, while Light-ASPP enhances the network’s ability to aggregate contextual information with minimal computational cost, making it suitable for large-scale applications. The dual-branch decoder explicitly decouples the learning of semantic segmentation and boundary prediction, ensuring more accurate boundary detection and crack continuity. The extensive experiments on multiple benchmark datasets demonstrate that BSA-Net consistently outperforms existing crack detection models, particularly in complex, noisy environments. The model achieves competitive performance in terms of segmentation accuracy, boundary clarity, and recall rates, particularly for fine-scale and weak contrast cracks. The results indicate that BSA-Net not only enhances the performance of crack segmentation in real-world conditions but also provides a scalable and reliable solution for automated infrastructure monitoring and defect detection.

Micro-crack detection on concrete surfaces is challenging because labeled micro-crack data are scarce, crack cues are extremely weak (often only a few pixels wide), and complex backgrounds (e.g., non-uniform illumination, shadows, and stains) degrade feature extraction; this study aims to improve both data availability and detection robustness for practical inspection. A Poisson image editing-based synthesis strategy is developed to generate visually coherent micro-crack samples via gradient-domain blending, and a Complex-Scene-Tolerant YOLO (CST-YOLO) detector is proposed on top of YOLOv10, following an “lighting decoupling–global perception–micro-feature enhancement” design. CST-YOLO integrates an Lighting-Adaptive Preprocessing Module (LAPM) to suppress illumination/shadow perturbations, a Spatial–Channel Sparse Transformer (SCS-Former) to model long-range crack topology efficiently, and a Small Object Focus Block (SOFB) to enhance micro-scale cues under cluttered backgrounds. Experiments are conducted on a 650-image dataset (200 real and 450 synthesized), in which synthesized samples are used only for training, and the validation/test sets contain only real images, with a 7:2:1 split. CST-YOLO achieves 0.990 mAP@0.5 and 0.926 mAP@0.5:0.95 at 139 FPS, and ablation results indicate complementary contributions from LAPM, SCS-Former, and SOFB. These results support the effectiveness of combining realistic synthesis and architecture-level robustness for real-time micro-crack detection in complex scenes.

During the process of remote microwave puncture ablation surgery in areas such as the liver and thoracic cavity, there are accompanying physiological movements due to the body's respiratory motion, as well as possible spasms and twitching of the body's muscles caused by factors like anesthesia and temperature. These physiological activities can easily lead to skin lacerations near the rigid percutaneous puncture ablation needle due to relative motion. To prevent this secondary harm to the surrounding skin caused by such movements, we designed a robot end-effector active respiration tracking method based on fuzzy variable admittance control of a robotic arm and sliding window half-wave analysis debounce algorithm. This method not only enables smooth zero-force tracking but also monitors real-time end-effector motion information to adjust the admittance parameters, thus avoiding the occurrence of potential oscillations. Finally, we conducted experiments by applying a random step-like oscillation to simulate spasms and twitching phenomena. The robot not only achieved zero-force tracking with forces less than 1.5N but also demonstrated excellent resistance to oscillations, showcasing remarkable flexibility and stability.

Existing studies on rehabilitation robots are generally devoted to robot-assisted active rehabilitation training, which is conducive to facilitating muscle and nerve regeneration. However, human–robot interaction (HRI) requires imposing a limit on the workspace within which the robot operates, so as to ensure patient safety. A safe admittance boundary algorithm for a rehabilitation robot is proposed based on the space classification model which works by constructing a virtual boundary for the HRI workspace in the control layer. First, point cloud isodensification is performed for the workspaces within which the human body and the robot operate. Next, the nearest neighbor density is determined for the point cloud, and the space classification model is built on this basis. Finally, the space classification model is integrated with admittance control to derive the safe admittance boundary algorithm, which can be used for safety control. This algorithm is then subjected to space verification experiments and out–of–bounds experiments using a dynamic arm simulator (DAS). As indicated by the experimental results, when the side length of the voxel grid for voxel center filtering is set to 0.06 m, the accuracy of space definition is 98.40%. The average maximum response time for out–of–bounds behaviors is 165.62 ms. The safe admittance boundary algorithm can impose reasonable virtual constraints on the robot workspace, thereby improving HRI safety.

Objective: Since Augmented Reality (AR) and 3D visualization have proven to be of great significance to the safety and effectiveness of surgical outcomes, will autostereoscopic 3D AR without glasses bring new opportunities for surgical navigation of laparoscopic surgery? Methods: We used the CPD-based deformation registration algorithm and the proposed virtual view generation algorithm, realizing a deformable autostereoscopic 3D AR navigation framework for laparoscopic surgery. The depth perception and user experience of the 3D AR navigation were evaluated compared with the 2D AR display using an in-vitro porcine heart and offline clinical partial nephrectomy laparoscopic images. Results: The autostereoscopic 3D AR allowed participants to have a more consistent spatial perception as well as a shorter measuring time than 2D AR with significant difference of p<0.05. It can also improve relative depth perception for smaller distance separation of objects < = 3.28 mm. However, the autostereoscopic 3D AR perceived a worse experience compared to 2D AR in the user experience. Conclusion: Autostereoscopic 3D AR shows a more efficient and robust sense of spatial scale than 2D AR with better potential to shorten the operating time and improve surgical outcomes than 2D AR, but image blur and distortion are issues that must be solved to improve the perception effect. High precise registration and high fluency visualization requirements could make autostereoscopic 3D AR navigation for soft tissue more challenging. Significance: Our work lays a theoretical foundation for the further development of laparoscopic surgical navigation.

A wall surface inspection robot mainly relies on the inertial measurement unit and global positioning system (GPS) signal during intelligent wall surface inspection. The robot may encounter incorrect positioning under a GPS-denied environment, easily triggering safety accidents. In order to obtain a path suitable for the safe work of the robot wall surface inspection robot in a GPS-denied environment, a global path-planning method for wall surface inspection robots was proposed based on the improved generic algorithm (GA). The influencing factor for GPS signal strength was introduced into the heuristic function in path planning for GA to address the aforementioned problem. Using the PSO algorithm, GA was initialized and the influencing term of GPS signal was introduced into the fitness degree function so as to achieve point-to-point path planning of vertical wall surface inspection robot. Path angle and probability of intersection and variation was taken into account for better path planning capability. Finally, the simulation experiments were performed. The generated path using the improved GA was found to avoid the blind area of the GPS signal. The algorithm proposed has a good performance with average convergence times of 35.9 times and an angle of 55.88° in simple environment. Contrary to the traditional GA and PSO algorithm, the method showed advantages in terms of the convergence rate, path quality, path angle change, and algorithm stability. The research presented in this article is meaningful and relatively sufficient. The simulation test is also quite convincing. The proposed method was proved to be effective in global path planning for a wall surface inspection robot.

Tracking and grasping a moving target is currently a challenging topic in the field of robotics. The current visual servo grasping method is still inadequate, as the real-time performance and robustness of target tracking both need to be improved. A target tracking method is proposed based on improved geometric particle filtering (IGPF). Following the geometric particle filtering (GPF) tracking framework, affine groups are proposed as state particles. Resampling is improved by incorporating an improved conventional Gaussian resampling algorithm. It addresses the problem of particle diversity loss and improves tracking performance. Additionally, the OTB2015 dataset and typical evaluation indicators in target tracking are adopted. Comparative experiments are performed using PF, GPF and the proposed IGPF algorithm. A dynamic target tracking and grasping method for the robot is proposed. It combines an improved Gaussian resampling particle filter algorithm based on affine groups and the positional visual servo control of the robot. Finally, the robot conducts simulation and experiments on capturing dynamic targets in the simulation environment and actual environment. It verifies the effectiveness of the method proposed in this paper.

A global path planning method is proposed based on improved ant colony optimization according to the slow convergence speed in mobile service robot path planning. The distribution of initial pheromone is determined by the critical obstacle influence factor. The influence factor is introduced into the heuristic information to improve the convergence speed of the algorithm at an early stage. A new pheromone update rule is presented using fuzzy control to change the value of pheromone heuristic factor and expectation heuristic factor, adjusting the evaporation rate in stages. The method achieves fast convergence and guarantees global search capability. Finally, the simulation results show that the improved algorithm not only shortens the running time of global path planning, but also has a higher probability of obtaining a global optimal solution. The convergence speed of the algorithm is better than the traditional ant colony algorithm.

Cardiovascular image registration is an essential approach to combine the advantages of preoperative 3D computed tomography angiograph (CTA) images and intraoperative 2D X-ray/ digital subtraction angiography (DSA) images together in minimally invasive vascular interventional surgery (MIVI). Recent studies have shown that convolutional neural network (CNN) regression model can be used to register these two modality vascular images with fast speed and satisfactory accuracy. However, CNN regression model trained by tens of thousands of images of one patient is often unable to be applied to another patient due to the large difference and deformation of vascular structure in different patients. To overcome this challenge, we evaluate the ability of transfer learning (TL) for the registration of 2D/3D deformable cardiovascular images. Frozen weights in the convolutional layers were optimized to find the best common feature extractors for TL. After TL, the training data set size was reduced to 200 for a randomly selected patient to get accurate registration results. We compared the effectiveness of our proposed nonrigid registration model after TL with not only that without TL but also some traditional intensity-based methods to evaluate that our nonrigid model after TL performs better on deformable cardiovascular image registration.