Different from most current studies using convolutional neural network (CNN), a deep learning detection method for steel plate surface defects based on Transformer is researched. This paper presents a process, network structure and detection method for steel strip surface defect detection. A Swin Transformer was used to extract hierarchical features in the detection system. Then feature pyramid networks (FPN) was used to fuse the above features to form multi‐scale feature maps, and region proposal network (RPN) was adopted to generate the generate region of interest (ROI) of defects. Finally, the ROI head was used to generate defect category information and its precise location. Here, ablation experiments were conducted to explore the impact of different backbone networks and the number of stages of Swin Transformer and FPN on target detection performance to prove the rationality and efficiency of the network. On the NEU‐DET dataset, the proposed algorithm achieved 81.1% mAP, which is higher than that of classic CNN detection methods such as Faster R‐CNN, SSD, Yolo v3 and RepPoints.

To overcome the limitations of current visual and laser SLAM methods in narrow, feature-repetitive, and lighting-variable degraded scenarios, we propose a robust real-time mapping and localization system utilizing semantic dimensional chains (SDCs). SDC is a novel ordered semantic instance map representation. During the mapping phase, our system efficiently integrates diverse object detection and 2D-SLAM algorithms, constructing a prior semantic map through Bayesian filtering. In the preprocessing phase, a semantic map optimization algorithm based on neighborhood homogeneity is applied to effectively eliminate semantic noise. For localization, SDCs combined with precise localization algorithms enable rapid global localization. The system also features a kidnapping detection mechanism for swift pose recovery. We rigorously evaluated the system’s performance in terms of robustness to lighting interference, global localization, and pose recovery in real-world degraded environments. Experimental results indicate that our localization system outperforms existing methods. Compared to a state-of-the-art vision-based localization algorithm, our semantic matching (SM) module exhibits superior resistance to lighting interference, resulting in a 15.38% increase in localization success rate (SR) and a 27.71% reduction in average localization time.

This article investigates the relocalization of robots in indoor environments. To achieve this, indoor corners were classified into eight distinct categories, and a two-dimensional grid map was constructed using various sensors. Deep learning technology was employed to extract semantic information from corners, and the Bayesian method was used to build a semantic corner map incrementally. Additionally, the class attributes and positional relationships for each corner were explored, thereby facilitating the establishment of a dimension chain for semantic corners. Furthermore, a fast and efficient method was introduced for retrieving this dimension chain. During the relocalization process, the dimension chain of semantic corners was utilized for initial positioning, followed by the application of improved adaptive Monte Carlo localization (AMCL) algorithm for precise localization. Through comparative analysis using AMCL and several state-of-the-art methods, superior performance in both localization success rate and real-time implementation was demonstrated. Finally, extensive relocalization experiments were conducted to validate the effectiveness of the proposed method.

This paper presents an innovative intelligent navigation and localization system designed for indoor dynamic environments, leveraging the Semantic Dimension Chain Knowledge Base Model (SDC-KBM). By modeling semantic maps, the system generates the Semantic Dimension Chain (SDC) and introduces the Semantic Dimension Chain Localization (SDCL) algorithm, enabling robust real-time localization in dynamic settings. The SDC is further modeled to build SDC-KBM, which consists of regional, instance, and operational layers. Based on this model, we propose the Semantic-Geometric Pattern-based Path Planning (SGPP) algorithm, which overcomes the low intelligence of traditional path planning methods and significantly enhances real-time performance. Additionally, a task rule-based semantic parsing algorithm interprets human instructions through SDC-KBM, allowing robots to adaptively navigate based on user intent and environmental semantics. Experimental results from real-world scenarios demonstrate that SDCL outperforms state-of-the-art localization algorithms, particularly in challenging corridor environments. Meanwhile, SGPP reduces processing time by 58.90% compared to Dijkstra and 38.49% compared to A-star.

To address the localization challenges in indoor dynamic and degraded environments such as long corridors with sparse and repetitive features, we present a semantic instance-based multisensor fusion localization (SIMSL) system. We introduce a multistrategy corner classification model that achieves accurate corner classification by integrating multisensor data. We combine Bayesian filtering with semantic segmentation to build a semantic instance map enriched with corner information and build a semantic dimension chain using a path exploration model. In terms of localization, we propose a robust semantic dimension chain matching algorithm, which, when integrated with scan-to-map, enables millisecond-level precise global localization even in dynamic and degraded environments. Additionally, our multisensor cumulative error compensation and kidnapping detection mechanisms effectively eliminate the cumulative errors from odometry and IMU while accurately detecting localization anomalies and restoring pose. Extensive localization experiments conducted in various real-world environments demonstrate that our proposed algorithm outperforms existing methods. Compared to a state-of-the-art algorithm, the global localization success rate improved by 12.51%, and the average localization time decreased by 52.19%.

Perception of the indoor environment is the basis of mobile robot localization, navigation, and path planning, and it is particularly important to construct semantic maps in real time using minimal resources. The existing methods are too dependent on the graphics processing unit (GPU) for acquiring semantic information about the indoor environment, and cannot build the semantic map in real time on the central processing unit (CPU). To address the above problems, this paper proposes a non-use GPU for lightweight indoor semantic map construction algorithm, named NGLSFusion. In the VO method, ORB features are used for the initialization of the first frame, new keyframes are created by optical flow method, and feature points are extracted by direct method, which speeds up the tracking speed. In the semantic map construction method, a pretrained model of the lightweight network LinkNet is optimized to provide semantic information in real time on devices with limited computing power, and a semantic point cloud is fused using OctoMap and Voxblox. Experimental results show that the algorithm in this paper ensures the accuracy of camera pose while speeding up the tracking speed, and obtains a reconstructed semantic map with complete structure without using GPU.