Zhenguo Nie

Endoscopic images are often degraded by uneven illumination, motion blur, and tissue occlusion, which obscure critical anatomical details and complicate surgical manipulation. This issue is particularly pronounced in single-port endoscopic surgery, where the imaging capability of the camera is further constrained by limited working space. While deep learning-based enhancement methods have demonstrated impressive performance, most existing approaches remain too computationally demanding for real-time surgical use. To address this challenge, we propose an efficient stepwise endoscopic image enhancement framework that introduces an implicit degradation representation as an intermediate feature to guide the enhancement module toward high-quality results. The framework further exploits the temporal continuity of endoscopic videos, based on the assumption that image degradation evolves smoothly over short time intervals. Accordingly, high-quality degradation representations are estimated only on key frames at fixed intervals, while the representations for the remaining frames are obtained through fast inter-frame propagation, thereby significantly improving computational efficiency while maintaining enhancement quality. Experimental results demonstrate that our method achieves an excellent balance between enhancement quality and computational efficiency. Further evaluation on the downstream segmentation task suggests that our method substantially enhances the understanding of the surgical scene, validating that implicitly learning and degradation representation propagation offer a practical pathway for real-time clinical application.

Rotation-invariant recognition of shapes is a common challenge in computer vision. Recent approaches have significantly improved the accuracy of rotation-invariant recognition by encoding the rotational invariance of shapes as hand-crafted image features and introducing deep neural networks. However, the methods based on pixels have too much redundant information, and the critical geometric information is prone to early leakage, resulting in weak rotation-invariant recognition of fine-grained shapes. In this paper, we reconsider the shape recognition problem from the perspective of contour points rather than pixels. We propose an anti-noise rotation-invariant convolution module based on contour geometric aware for fine-grained shape recognition. The module divides the shape contour into multiple local geometric regions (LGA), where we implement finer-grained rotation-invariant coding in terms of point topological relations. We provide a deep network composed of five such cascaded modules for classification and retrieval experiments. The results show that our method exhibits excellent performance in rotation-invariant recognition of fine-grained shapes. In addition, we demonstrate that our method is robust to contour noise and the rotation centers.

Magnetically controlled spinal growing rods, used for treating early-onset scoliosis (EOS), face a critical clinical limitation: insufficient distraction force. Compounding this issue is the inherent inability to directly monitor the mechanical output of such implants in vivo, which challenges their safety and efficacy. To overcome these limitations, optimizing the rotor’s maximum torque is essential. Furthermore, defining the “continuous rotation domain” establishes a vital safety boundary for stable operation, preventing loss of synchronization and loss of control, thus safeguarding the efficacy of future clinical control strategies. In this study, a transient finite element magnetic field simulation model of a circumferentially distributed permanent magnet–rotor system was established using ANSYS Maxwell (2024). The effects of the clamp angle between the driving magnets and the rotor, the number of pole pairs, the rotor’s outer diameter, and the rotational speed of the driving magnets on the rotor’s maximum torque were systematically analyzed, and the optimized continuous rotation domain of the rotor was determined. The results indicated that when the clamp angle was set at 120°, the number of pole pairs was one, the rotor outer diameter was 8 mm, the rotor achieved its maximum torque and exhibited the largest continuous rotation domain, while the rotational speed of the driving magnets had no effect on maximum torque. Following optimization, the maximum torque of the simulation increased by 201% compared with the pre-optimization condition, and the continuous rotation domain was significantly enlarged. To validate the simulation, a rotor torque measurement setup incorporating a torque sensor was constructed. Experimental results showed that the maximum torque improved from 30 N·mm before optimization to 90 N·mm after optimization, while the driving magnets maintained stable rotation throughout the process. Furthermore, a spinal growing rod test platform equipped with a pressure sensor was developed to evaluate actuator performance and measure the maximum distraction force. The optimized growing rod achieved a peak distraction force of 413 N, nearly double that of the commercial MAGEC system, which reached only 208 N. The simulation and experimental methodologies established in this study not only optimizes the device’s performance but also provides a viable pathway for in vivo performance prediction and monitoring, addressing a critical need in smart implantable technology.

The carbon concentration and hardness of a part’s surface can be enhanced through the carburizing-quenching process. The mechanical properties are affected by the carbon concentration distribution in the surface layer, which directly determines wear resistance and fatigue life. Carbon concentration distribution can be calculated using the FEA (Finite Element Analysis) method with an error no greater than 10% for the heat treatment engineer. However, predicting the results using machine learning method is more efficient and easier. In this paper, the carburizing numerical model is established for a 20Cr2Ni4A cylindrical part. The microstructures and hardness gradient were tested to verify the accuracy of the simulation model for the furnace samples after the carburizing process. By establishing an accurate numerical model of the carburizing process, multiple sets of process parameter combinations were designed for three different shapes: square, circle, and trapezoid. Large-scale calculations were performed, resulting in a total of 530,226 sets of samples obtained. Finally, a parameter adaptive BPNN (Backpropagation Neural Network) method based on mean shift clustering and slime mould algorithm is proposed for rapid prediction of carbon concentration distribution on the surface of parts after carburizing. By comparing the prediction accuracy and training time with other methods, the feasibility and effectiveness of the method are verified.

With an increasing number of mechanical components produced in the production pipeline, the need to classify past and new elements efficiently has been increasing. However, past methods of classifying elements have relied on traditional methods that take a long time and lack extensibility. By incorporating specific geometric local features, graph neural networks, and attention between point clouds to include local and global features, we propose a new framework called Mechanical Component Classifier (MC Classifier), which can classify components efficiently with a fast inference time and can be easily extended to classify other elements. We benchmark the performance of MC Classifier against state-of-the-art models and demonstrate its competitive potential in 3D mechanical component classification. Our findings suggest that MC Classifier has significant potential to advance 3D mechanical component classification. Our findings show that MC Classifier has significant potential to advance 3D mechanical component classification and efficient and affordable methods to streamline industrial pipelines.