The external parameter calibration of LiDAR and camera is the core link of multimodal perception fusion technology, and its accuracy directly determines the effectiveness of cross modal information fusion. This article proposes a method for high-precision calibration of laser radar and camera extrinsic parameters using a checkerboard calibration board based on spatial registration theory. In response to the difficulty of feature extraction caused by the sparsity of LiDAR point clouds, this method enhances feature information by accumulating multi view point clouds and combines filtering algorithms to improve feature saliency; A standardized experimental platform was built based on ROS and Matlab, utilizing the reflection intensity distribution pattern of the checkerboard point cloud to achieve accurate extraction of its three-dimensional corner coordinates. At the same time, the corresponding two-dimensional checkerboard corner points are detected in the camera image, and the relative pose transformation matrix between the LiDAR and the camera is solved by establishing an accurate matching relationship between the two-dimensional and three-dimensional corner points. The experimental results show that this method has high-precision calibration results and good repeatability, which can provide a reliable extrinsic basis for multi-sensor fusion perception systems.

Due to the complex downhole environment in oil and gas wells, electromagnetic wave wireless transmission suffers from high attenuation, multipath interference, and low signal-to-noise ratio, making long-distance and reliable communication difficult. To address these challenges, a composite modulation method combining Differential Phase Shift Keying (DPSK) and Forward Error Correction (FEC) is proposed. A downhole channel model incorporating path loss, multipath effects, and noise interference was established, and a transmitting–receiving hardware system was designed. Signal transmission characteristics in different media were experimentally investigated to analyze electromagnetic signal propagation behavior. The encoding and decoding performance of the DPSK–FEC scheme was evaluated through MATLAB simulations, focusing on the phase disturbance resistance of DPSK and the error correction capability of FEC. The coding parameters were further optimized to match downhole channel characteristics. The non-coherent demodulation mechanism of DPSK provides robustness against phase noise and frequency drift, while FEC introduces redundancy to detect and correct transmission errors, significantly improving system reliability. Experimental results show that the proposed system can operate stably at a depth of 1000 m and an environmental temperature of 75°C, enabling reliable bidirectional transmission of downhole data over long distances.

Currently, more and more image recognition technologies are being used in AI, and these AIs are gradually being applied in productivity scenarios. This article talks about the performance of the two mainstream AI ChatGPT5 and Qwen VL-Max in recognizing abnormal images. This research divides the abnormal images into four categories: Anatomical Anomalies, Physical Law Violations, Functional and Contextual Incongruities, Scale and Proportion Paradoxes. It is hoped that through this research, we can discover whether different AIs have varying perceptions of abnormal images generated by divergent models, in order to determine how differences in the training dataset impact the model's performance. Through three progressively in-depth questions for each image, researchers found that in most questions, the difference between the two was not significant, and both could identify the issues in the images. However, regarding anatomical anomalies that most mainstream models struggled to detect, ChatGPT, although providing incorrect answers, reflected on them and proactively requested a comparison with normal structural models. Researchers hope that this result will provide insights for developers.

Alan Turing's machine model of the universal machine model laid the theory and foundation of current computers, and layers can be regarded as an essence of computer science of current generation computers, averting the ubiquity of the universal computing; however, gradually, against Moore’s law and stepping into higher speed-computing needs, betraying the shortcomings of Turing’s model, are come up a new structure paradigm, namely specific domain architectures. Domain architecture is a style of software development which can extracts the complex business logic and build it out as a pure ‘domain model’, one that is independent of technical solutions. It is a bit like preparing an exact business blueprint of a software system: the software code may truly reflect business concepts and business rules. By defining clear boundaries and responsibilities, the domain architecture makes the system highly understandable, maintainable and capable of dynamically responding to changes in frequent business requirements. This article argues that people are entering a "post-Turing era", and people gradually turn from the pursuit of universality to pursuit of efficiency and accuracy. And that success of domain-specific architectures is no mere technology nor is it a new form rather than Turing’s form as represented by GPUs. Moreover, the future of computing will, sooner or later, be a heterogeneous era dominated by this new form.

Lung cancer is one of the most common and deadly types of cancer worldwide. While computed tomography (CT) of the chest is considered the gold standard for early detection, manual evaluation faces two major challenges: Micronodules are often overlooked, and radiologists suffer from reduced efficiency due to overwhelming daily case numbers and persistent fatigue. To solve these problems, we employ the LUNA16 dataset (including 888 CTs, 1186 annotated nodules and more than 550K candidate samples) to build an integrated experimental platform and systematically compare four kinds of detection methods: traditional manual features, 2D deep learning, 3D deep learning and transformer model. The evaluation metrics include detection accuracy (AUC-ROC curve and recall), ability of micronodule detection, robustness on limited training data (10% / 25% / 50% data amount) and computation efficiency (parameter scale and frame rate). The experimental results show that 3D Swin Transformer T has the best performance. Compared with the traditional method HOG+Random Forest, test AUC-ROC reaches 0.948 and the recall of micro-nodule is 89.7%, which is 34.4% improvement. 3D ResNet-50 has a good robustness in small sample situation (AUC could still be 0.902 when using 10% data for training), and 2D ResNet-50 could get a balance between fast speed and accuracy (85 fps). The experimental results provide support for the selection of model in clinical computer-aided system for lung cancer detection.

At present, society's awareness of environmental protection has been enhanced, and the demand for simultaneous development and protection of environmental purification and sustainability is increasing. As a clean energy source, wind power generation can well replace a part of fossil energy and prevent the inevitable pollutants caused by the use of fossil energy. However, the generators that wind power depends on do not have the ability to migrate, so the location of wind turbines is particularly important. It can not only occupy the potential residence of human beings, but also provide a large and stable amount of power generation. In this paper, three different machine learning models are used to explore the impact of a number of geographical factors on wind speed and to predict the future trend of wind speed. By comparing the three different models, the most suitable model with the highest fitting rate is XGBoost model, and its corresponding is 0.92. The scatter diagram of wind speed prediction and longitude and latitude produced in this paper can further assist in the location of wind turbines to maximize energy efficiency.

High-performance near-infrared photodetection is of great importance for applications in optical communication and imaging. Here, a vertically stacked WS2/MoTe2 van der Waals heterojunction with dual hBN encapsulation is constructed. Benefiting from interface purification and synergistic band structure modulation, the device exhibits significantly enhanced photoconversion efficiency and an extended spectral response range. A high responsivity of 13021 A W-1 and a detectivity of 5.39 × 1014 Jones are achieved at 405 nm. In the near-infrared region, responsivities of 586 A W-1 and 2292 A W-1 are obtained at 808 nm and 1064 nm, respectively, with a maximum detectivity of 9.49 × 1013 Jones. Stable time-response characteristics are maintained across all measured wavelengths. These results demonstrate that interfacially purified van der Waals heterostructures provide an effective strategy for high-performance two-dimensional near-infrared photodetectors.

To address the challenges of complex coupling of process parameters and limited accuracy of traditional predictive models in sintered ore production, this study proposes an intelligent optimization system framework integrating multiple algorithms. First, through data governance and feature engineering, the XGBoost gradient boosting tree algorithm is used to extract key process features and establish a nonlinear correlation model. Second, t-SNE high-dimensional data dimensionality reduction technology is introduced to enable visual diagnosis and anomaly detection of the sintering process patterns. Finally, the KNN clustering algorithm is innovatively integrated with the TOPSIS multi-objective evaluation model to construct a dynamic grading and preferential decision-making framework for process plans. Experimental analysis shows that this system can significantly enhance the stability and energy efficiency of the production process, providing technical support for the intelligent transformation of the steel industry.

Accurate and real-time traffic sign detection is a critical component of environmental perception for Advanced Driver-Assistance Systems (ADAS) and autonomous vehicles, playing a vital role in ensuring driving safety. However, current mainstream lightweight object detection models often encounter performance bottlenecks when dealing with challenges prevalent in real-world road scenes, such as small objects and complex backgrounds, primarily due to the loss of feature information. To address this issue, this paper conducts an in-depth application and performance evaluation of an improved YOLOv11-based lightweight detection algorithm. The algorithm is characterized by two key features: first, it enhances the perceptual capabilities for tiny objects by introducing a higher-resolution P2 feature layer; second, it adopts an integrated fusion detection module, termed FPN-P2 in this paper, which applies a "shallow feature direct-path, deep feature fusion" strategy to efficiently integrate multi-scale features. Extensive experiments on the public TT100K dataset demonstrate that the optimized configurations of this architecture achieve significant breakthroughs in both performance and efficiency. Specifically, its performance-optimal configuration achieves a mAP@0.5 of 76.3%, substantially surpassing baseline models like YOLOv8n. Meanwhile, its efficiency-optimal configuration attains a high accuracy of 68.8% while having fewer parameters than the baseline model. This research not only provides and validates highly competitive solutions for high-precision lightweight traffic sign detection but also, through intensive ablation studies, offers valuable empirical insights into model architecture selection and optimization for specific tasks.

The rapid advancement of intelligent driving technology has increased communication between vehicles, external networks, cloud platforms, and other vehicles, significantly elevating cybersecurity risks. Since intelligent driving systems rely heavily on in-vehicle networks and wireless communications, they are vulnerable to various network attacks. Traditional security measures often fail to meet the stringent real-time and accuracy requirements of such systems, making it urgent to develop efficient and intelligent intrusion detection systems to ensure safe and stable operation. This paper begins by analyzing the network architecture and security requirements of intelligent driving systems, and reviews mainstream network intrusion detection technologies—including feature matching, anomaly detection, and methods based on machine learning and deep learning. By comparing the strengths and weaknesses of different approaches, it highlights the challenges related to real-time performance and accuracy that conventional detection techniques face in intelligent driving environments. The study also discusses future trends in intelligent driving cybersecurity, emphasizing the importance of integrating AI technology with cybersecurity to enhance intrusion detection capabilities. Finally, the paper proposes research directions such as multi-level defense systems and cross-domain collaborative defense, offering both theoretical and practical support for securing intelligent driving systems.