The U-Net architecture, with its skip connections and encoder-decoder structure, plays a pivotal role in medical image segmentation. While various U-Net-based models attempt to enhance skip connections through basic operations (e.g., arithmetic addition/subtraction) or attention mechanisms, they overlook the distinctive characteristics of features at different encoding stages. This paper proposes MFIN - a novel medical image segmentation network that incorporates a Multi-scale Fusion Module (MFM) to process feature maps from different encoding stages through diversified operations in skip connections. This mechanism effectively leverages multi-level image details and semantic information to assist the decoder in generating precise segmentation results. Additionally, we introduce an Attention Enhancement Module (AEM) in the decoder phase, where a coarse segmentation map produced by preliminary decoding guides subsequent feature reconstruction through an attention refinement process. Extensive experiments across six datasets demonstrate the superior performance of MFIN, achieving an average mDSC of 89.84% and significantly outperforming the baseline U-Net model.
In response to the inspection demands in the complex environment of train bottom, this paper proposes a virtual teaching system based on digital twin technology to improve deployment efficiency and safety of train inspection robots. The virtual teaching system utilizes a five-dimension (5D) digital twin system to seamlessly integrate the virtual and physical domains of the train inspection robot via the Isaac Sim and the Robot Operating System (ROS). It features two test modes, Software-in-the-loop (SIL) and Hardware-in-the-loop (HIL), enabling path planning and motion control to be tested in a photo-realistic environment, with optimized teach-in data deployed to the physical robot. The system can effectively optimize the inspection process, improve automation, and reduce risks and costs during deployment.
This study proposes a method for automatic audit report generation and compliance analysis based on the BERT model, aiming to solve the problems of low efficiency, high labor costs, and insufficient compliance detection accuracy in traditional audit work. By combining the powerful semantic understanding ability of the pre-trained language model, this study achieves high-quality generation of audit reports and uses classification tasks to accurately identify potential risk points in the text. The experiment uses multiple public datasets to evaluate the performance of BERT in text generation quality (BLEU Score) and compliance detection (Accuracy, F1-score). Compared with models such as GPT, T5, and LSTM, BERT performs best in all indicators. At the same time, through robustness testing, the performance of the model in the face of text noise or tampered data is analyzed. The results show that although noise has a certain impact on model performance, BERT's overall performance is still better than that of other comparison models. This study also explores the impact of different fine-tuning strategies on model performance and analyzes its adaptability and limitations in the audit field. The research findings demonstrate that the BERT-based methodology not only effectively enhances the quality of the automated generation of audit reports but also substantially improves the accuracy of compliance analysis. This advancement offers novel insights and serves as a guiding framework for the development of intelligent audit technology.
Recent advancements in Novel View Synthesis (NVS) have primarily focused on Low Dynamic Range (LDR) images, with methods like Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) improving performance. However, research on using RAW images, which offer a wider dynamic range and greater detail, is limited. This study investigates the use of noisy RAW and LDR images in NVS tasks, comparing RawNeRF and 3DGS. The results show that RAW images provide superior reconstruction quality, particularly in noisy conditions. While RawNeRF excels in noise suppression and detail recovery, it is slower, whereas 3DGS offers faster processing, making it more suitable for real-time applications. This work demonstrates the necessity of RAW data for high-quality synthesis and highlights the potential of 3DGS for future real-time applications.
The Rod Cluster Control Assembly (RCCA), a nuclear safety-critical component, is susceptible to wear, swelling, and other forms of damage during prolonged service, posing substantial risks to the safe and stable operation of nuclear reactors. While the functional importance of control rods is selfevident, research on their health status remains limited in both domestic and international contexts. This paper summarizes and analyzes damage patterns based on multiple in-service inspection data from RCCAs in six reactor units at a Chinese nuclear power plant, aiming to optimize control rod assembly management. The study revealed that: (1) Single-wing rods exhibit significantly higher wear probability compared to double-wing rods, with particular attention required for single rods numbered G3, A5, C7 and E1; (2) Control rod damage predominantly concentrates in the lower end plug region approximately 200 mm from the bottom; (3) Wall thickness shows regular thinning patterns with extended service time, and preliminary remaining service life assessment can be conducted through box plot trend analysis of wall thickness reduction.
