The inherent sparsity and irregular geometric structure of point clouds easily lead to feature deviations among different local regions. On the one hand, most existing approaches that directly acquire semantic features from different local regions suffer from reduced accuracy and stability while failing to exploit more discriminative information. On the other hand, relying solely on raw coordinates to implicitly represent geometric relationships cannot adequately capture explicit Euclidean geometric priors of point clouds for richer spatial encoding. To address the above issues, we propose a locally adaptive geometry-aware network. Its feature affine decoupling module employs semantic centroid-guided affine transformations to derive robust semantic features while enhancing discriminative feature extraction through a dual-path decoupling mechanism. The geometric relation enhancement module learns geometric priors within local regions to capture the local shape and geometric details of each point and transforms these into higher-order representations of spatial relations to explicit reasoning about point spatial layouts. The local adaptive geometry-aware module interactively captures geometric and semantic features within the point cloud through a dual-path architecture to establish precise spatial relationships and achieve geometry-aware perception of local shapes. The model was tested and validated on the S3DIS and ScanNetV2 datasets, achieving mean Intersection over Union (mIoU) scores of 77.9% and 71.8%, respectively. In addition, experiments were also conducted on the SemanticKITTI dataset. The validation results indicate that the method exhibits excellent segmentation performance in large-scale point cloud semantic segmentation scenarios.

Currently, existing point cloud semantic segmentation methods do not fully exploit surface geometric features. In particular, the depiction of object boundaries and the transition areas of curved surfaces is rather rough. On the other hand, the neighbourhood aggregation mostly follows a single strategy, making it difficult to simultaneously take into account the context and fine-grained differences and ignoring local details. To address these issues, this paper proposes a geometry-enhanced adaptive local feature aggregation network (GALA-Net). First, a geometric information embedding (GIE) module is introduced, which extracts pseudo-normal vectors and pseudo-curvatures of local point cloud regions as geometric priors, and incorporates multi-frequency sine-cosine encoding to capture multi-scale spatial relationships, yielding enhanced local geometric representations. Then, an adaptive feature fusion (AFF) module dynamically allocates fusion weights between semantic and geometric features, thereby alleviating channel coupling and neighbourhood noise amplification caused by simple concatenation. Next, a dual-path adaptive attention aggregation (DAAA) module jointly models semantic and positional attention and adaptively fuses them with max-pooled features to improve the robustness of local aggregation. In addition, a self-enhanced attention encoding (SEAE) module is designed to expand the feature representation space by extracting features through independent mapping branches and fusing them in a residual manner. The proposed model is evaluated on the S3DIS and ScanNetV2 datasets, achieving mIoU scores of 78.0% and 71.6%, respectively, which demonstrates its strong segmentation performance on indoor scenes.

Steel surface defect detection is a critical aspect in ensuring the quality of steel production. However, due to the small size, irregular shapes, and complex backgrounds of defects, existing detection algorithms face significant challenges in feature extraction and recognition, struggling to balance high accuracy with real-time performance. To address these issues, this study proposes an improved YOLOv8 model–CSG-YOLO, which integrates deformable convolution, a Simplified Cross-stage Partial Spatial Pyramid Pooling Fusion (SimCSPSPPF) module, and a Global Feature Pyramid Network (GFPN). The model aims to efficiently extract robust features of minor and irregular defects while meeting the stringent demands of industrial environments for real-time high-performance computing. First, we replace the standard convolutions in the last three C2f modules of the original YOLOv8 backbone with deformable convolution, enabling the model to adapt to the geometric deformations of irregular defects and significantly enhancing the flexibility of feature extraction. Second, we introduce the SimCSPSPPF module to replace the original SPPF module, enhancing the diversity and effectiveness of features through cross-layer feature fusion and multiscale pooling. Finally, we adopt the GFPN structure in place of the original PAFPN, enabling cross-scale global feature interaction to integrate richer contextual information. Experiments on the Northeastern University (NEU-DET) dataset show that CSG-YOLO achieves an mAP@0.5 of 76.1

To address the low segmentation accuracy of most deep learning-based point cloud networks-particularly the mis-segmentation of transition regions-which stems from a lack of interpoint interaction within local areas, substantial information loss from max-pooling operations, and the forced aggregation of heterogeneous semantic features during feature refinement, we propose a network named PointNeXt-CNM. In the feature optimization stage, a channel-level self-attention mechanism is used to adjust the point cloud features of local regions to enhance the information interactions between points within local regions. Then, a reverse aggregation method is used to integrate the adjusted local region point cloud features to enhance the information interactions between different local regions, which enhances the point cloud feature representation of the local regions and expands the model's receptive field. In the feature aggregation stage, mixed pooling (combining channel-level attention pooling and max pooling) is used to improve the proportion of critical information and preserve the distinct features of local regions to effectively alleviate the problem of information loss. In comparison experiments, compared with those of the PointNeXt-S model, the mIoU and OA of the PointNeXt-CNM model were improved by 3.1% and 2.02%, respectively, on the S3DIS dataset; the mIoU of the PointNeXt-CNM model was improved by 0.4% on the ShapeNetPart dataset; and the mAcc and OA of the PointNeXt-CNM model were improved by 1.54% and 0.16%, respectively, on the ModelNet40 dataset. The results show that the PointNeXt-CNM model achieved higher segmentation accuracy than various mainstream networks did, especially in transition regions.

Point cloud semantic segmentation is a crucial technique for accurately perceiving and understanding 3D scenes. However, due to the inherent sparsity and irregularity of point cloud data, it remains a highly challenging task. Most existing methods learn geometric features by directly relying on point coordinates, but they fail to exploit the intrinsic geometric structure of point clouds fully and lack effective interaction between features and geometric relationships, which leads to ambiguous local spatial perception. To address this issue, we propose a Hybrid Feature Affine and Geometric Context Aggregation Network (HGCA-Net). The network adaptively aligns point features via Hybrid Feature Affine (HFA), alleviating problems caused by sparsity and uneven density, and employs a novel Local Spherical Representation (LSR) to capture the non-Euclidean properties of point clouds. Based on this, we design a Local Geometric Relation Attention (LGRA) mechanism to enable interaction between point cloud features and geometric information, effectively learning local point cloud details. Finally, a Geometric Context Feature Aggregation (GCFA) module is introduced to enlarge the receptive field and reduce information loss. Experimental results show that HGCA-Net achieves 72.1

Because variability index CFAR (VI-CFAR) and switching variability index CFAR (SVI-CFAR) detectors are affected by the number and location of interference targets and the number of clutter reference cells, a nonhomogeneous environment cannot be recognized accurately, resulting in a detection performance decline in a nonhomogeneous environment. This paper proposes an adaptive censoring CFAR detector (AC-CFAR), which first calculates the position of the transition and then identifies whether the subreference window starting from the transition is in a homogeneous environment. Then, based on the position of the transition and whether the subreference window is in a homogeneous environment, an appropriate method was selected from CA-CFAR, GO-CFAR and OS-CFAR to calculate the detection threshold. Monte Carlo simulation results show that the detection performance of AC-CFAR is consistent with that of VI-CFAR and SVI-CFAR, and near that of CA-CFAR in a homogeneous environment, but its performance is better than that of VI-CFAR and SVI-CFAR in a multitarget environment with a large number of interference targets. In particular, the number of interference targets is uneven on both sides of the cell being tested. In a clutter edge environment with less clutter, the false alarm rate of AC-CFAR is marginally lower than that of VI-CFAR and SVI-CFAR.

Road damage detection plays a vital role in road safety maintenance. Minor road damage cannot often be identified efficiently because of the challenge of extracting damage information, which frequently lacks minor information. In addition, due to the irregular shape of alligator road damage, the performance of existing damage detection algorithms is seriously reduced. To solve the above problems, we propose an improved YOLOv5 model called YOLOv5-MCD for road damage detection. First, a multiscale dilated convolution module combined with an attention mechanism is proposed and used to replace the spatial pyramid pooling fast module in the neck of the original YOLOv5 model. Multiscale dilated convolution with a larger receptive field can be applied to obtain more damage feature information at different scales, and the attention mechanism aggregates the spatial and channel feature information of road damage images, allowing the model to pay more attention to damaged areas. Second, the cross-layer feature fusion method is proposed to merge more road damage features to enhance the original damage feature information. Finally, deformable convolution is used to improve the adaptability and performance of our method for complex alligator damage. Our method was tested and validated by the China-MotorBike RDD2022, and the F1-score and mAP@0.5 of our model were 93.5% and 95.5%, respectively. The experimental results show that our method can efficiently identify minor and alligator road damage and achieve better detection performance. (c) 2024 SPIE and IS&T

In order to solve the problem that the detection performance of the radar target detector decreases badly in non-homogeneous environments. Based on the actual echo clutter distribution, a Multiple Switching CFAR detector (MS-CFAR) is proposed. Firstly, the maximum reference units in the left and right reference sub windows are removed by censoring threshold. Then, compare the number of remaining reference cells in the left and right reference windows and variation index with the corresponding threshold respectively, and select the appropriate reference cells to accurately estimate the background noise power. Compared with the simulation and analysis results of other detectors, the proposed detector had the best detection performance and stability in multi-interfering targets, clutter edge and other non-homogenous environments. The results show that the proposed detector still has a good detection performance in non-homogeneous environments.

针对经典霍夫车道线检测方法实用性较差,无法准确区分车道线和路沿与应用道路场景简单等问题,提出了一种基于消失点和颜色过滤器的车道线检测算法,不仅提高车道线检测的准确率,而且能够应用较复杂行车场景;首先,对行车视频连续五帧图像进行预处理,获取行车环境下车道线消失点位置,能够自适应选取行车环境图像的感兴趣区域(Region of Interest,ROI);然后,对ROI图像根据车道线颜色特征进行过滤得到二值图像,获取二值图像中所有连通区域质心和倾斜角等数据,通过结合消失点特征和角度阈值进行限制,筛选记录符合车道线特征连通区域的数据,接着分割较大区域获取更多质心点,识别漏检符合车道线特征的区域质心点;最后,对获取的质心点使用最小二乘法进行拟合并标识车道线;实验结果表明:算法能够在多场景道路上快速准确的检测出车道线,与经典霍夫算法进行仿真比较,算法具有一定的鲁棒性和实时性.

In view of the problem of high false alarm rate of OS-CFAR in clutter edge environment, proposes a CFAR algorithm based on ordered statistics and maximum reference cell selection. Firstly, the reference window is divided into several sub-reference windows. Then, the traditional OS-CFAR strategy is adopted to select reference cells in each sub-reference window. Finally, the reference cell with the highest background noise power is selected from the selected reference cells and multiplied by the threshold factor to obtain the detection threshold. Simulation results show that the detection performance of the improved CFAR is slightly lower than that of OS-CFAR in homogeneous environment and multi-target environment. However, in clutter edge environment, especially at false alarm peak, the performance of the improved CFAR is significantly better than that of OS-CFAR.

工程教育专业认证是国际通行的工程教育质量保障制度,实现工程教育国际互认和工程师资格国际互认的重要基础.本校计算机科学与技术专业的课程以工程教育认证的理念为基础,对专业课程进行了一系列改革.本文以计算机科学与技术专业基础课《高级语言程序设计》为例,介绍计算机专业的相关课程落实工程认证"以学生为中心""成果导向""持续改进"的三大教育理念的具体举措.

针对非计算机专业开设的"计算机软件技术基础"课程在教学中存在的问题,提出了案例教学、信息化教学与实践教学相结合的教学方法,并在其基础上对评价体系进行了改革,以突出"学生为主体"的教育理念,对近两届的学生成绩进行了统计比较,从分析可得,新的教学方法有效地促进了学生的学习积极性与参与性.

In this paper a safety distance automatic control model (SDACM) and algorithm are introduced to solve the problem that the traditional vehicle safe distance control models have poor safety distance control. The algorithm uses road recognition and learning-aided module to enhance the self-adaptability for different brake system features, road conditions and the drivers. Based on the vehicle speed and braking performance, the proper period closed-loop safe distance control model are introduced to improve estimation and control accuracy of safety distance. The algorithms have been applied on Chery intelligent autonomous vehicle, through simulation and test under complex urban road condition, the minimum safe distance is maintained within 1.0–2.0 m, control accuracy is ±0.35 m. The results show that the control model and algorithm achieve much better control accuracy in complex urban conditions and effectively improve driving safety, comfort and road traffic efficiency.

针对当前真空助力主动制动系统助力保持时间短、位置控制精度低和控制算法实用性差的问题,提出了一种改进的系统及其控制算法.在真空腔、工作腔和大气之间增加带自锁功能的电磁阀,通过控制电磁阀实现主动助力控制,同时实现电磁阀掉电状态下的助力保持.通过增加的机械结构实现模式自动切换,保证人工制动的有效性.结合基于传感器距离反馈的安全距离和预制动控制算法,缩短了系统响应时间,提高了系统的实用性.系统及其控制算法成功应用于某电动智能车自主跟车系统中.经过测试,在城市工况下,自主跟车时最小安全距离控制在1.00～2.00 m范围.结果 表明:该系统可以满足车辆主动制动行车和驻车制动需求,能保证制动的安全性,同时提高行车安全性,实现任意长时间保压,响应快;控制算法保证了制动平顺性,同时提高了行车安全性和道路行车效率.

针对智能车辆自动紧急制动和自主跟车时安全距离控制稳定较差的问题,提出了一种最小安全车距控制算法,基于汽车制动距离分段精确估算模型,针对车辆制动过程的不同阶段,采用不同的制动距离估算模型,准确计算制动距离.尤其是初始制动距离的精确估算,可以准确给出实施制动的最佳时机.再结合周期安全距离闭环控制算法,实现了最小安全车距的精准控制.该模型及算法已应用于奇瑞智能驾驶自动紧急制动和自主跟车系统中,经过5种初始车速,3000次的实际工况试验,最小安全车距均保持在1.0～2.0 m内,控制精度<±0.5 m.结果表明:所提算法能够精确控制最小安全车距,尤其是针对前方静止目标的工况下,能够保证驾乘舒适性,同时有效提高车辆行驶安全性和道路行车效率.

针对当前电子液压主动制动系统保压时间短、响应慢和控制算法实用性差的问题,提出了一种改进的系统及其控制算法.在主缸和ABS或ESP之间的双管路上分别增加了两个常闭增压阀、减压阀和自锁电磁阀,取消了梭阀,使系统保留了双管路安全设计,实现了掉电保压,同时保证了系统在人工制动模式下的有效性.通过高压储能、双路增压和预制动,缩短了系统响应时间.采用基于距离和减速度的双闭环控制,提高了控制算法的实用性.经过测试,双管路上10 Mpa建压时间为170ms,控制精度±0.15 Mpa;9m/s2减速度响应时间为180ms,调节精度±0.1 m/s2;最小安全距离控制在1～2 m内.结果 表明,该系统可以实现任意长时间保压,并且响应较快.控制算法既保证了制动平顺性,同时也提高了行车安全性和道路行车效率.

为了解决传统分布式VI-CFAR(variability index-constant false alarm rate)和OS-CFAR(order statistic constant false alarm rate)算法在多杂波干扰环境下检测性能严重下降的问题,提出了一种分布式IVI-CFAR(improved VI-CFAR)检测算法.在VI-CFAR的基础上,引入双参考窗和双参量,通过各子参考窗参量与对应门限的比较,判断子参考窗中参考单元的分布情况,从而选择相应的参考单元集和检测器估计背景噪声功率.经过仿真对比,在单杂波和多目标干扰环境下,分布式IVI-CFAR算法检测性能接近于分布式VI-CFAR和OS-CFAR算法;在多杂波干扰环境下,分布式IVI-CFAR算法检测性能优于分布式VI-CFAR和OS-CFAR算法,采用"OR"融合规则时,检测性能最优.结果表明,提出的分布式IVI-CFAR算法在非均匀噪声环境下,均具有较优的检测性能.

为了解决有序统计恒虚警(order statistic constant false alarm rate,OS-CFAR)、有序统计最大选择恒虚警(order statistic greatest of-constant false alarm rate,OSGO-CFAR)和有序统计最小选择恒虚警(order statistic smallest ofconstant false alarm rate,OSSO-CFAR)检测算法在非均匀噪声环境下检测性能严重下降的问题,基于威布尔分布模型和模糊量化的软决策方法,提出了一种加权有序统计量的模糊恒虚警(weighted order statistic and fuzzy rules constant false alarm rate,WOSF-CFAR)检测算法。通过计算Leading和Lagging子窗口对应的模糊隶属函数值,采用代数积、代数和、最大选择和最小选择4种融合规则对2个子窗口的模糊输出量进行融合,并与比较门限进行比较判别目标有无。仿真表明,提出的检测方法与OSGO-CFAR,OSSO-CFAR算法相比,在均匀噪声、杂波边缘干扰和多目标干扰环境下均具有较好的检测性能,尤其是采用代数积融合规则时,检测性能最优,且提出的检测算法在均匀噪声环境下也具有最佳的检测性能。

为了解决传统的目标检测算法在非均匀噪声环境下检测性能严重下降的问题,提出了一种基于双剔除门限的Switching-CFAR（switching-constant false alarm rate based on dual censoring thresholds,DCS-CFAR）目标检测算法。基于参考窗参考单元样本期望值和测试单元,得到双重剔除功率比较门限。通过双重比较,剔除参考窗中极大值参考单元,根据剩余参考单元数,选择合适的参考单元来估计背景噪声功率,并得到功率检测门限。在Matlab环境下,通过蒙特卡洛方法和Swerling II模型对DCS-CFAR目标检测算法的关键参数,以及在各种仿真环境下与其它目标检测算法的检测性能进行了仿真对比分析,DCS-CFAR目标检测算法在均匀背景噪声下,检测率为98.76%,接近于CA-CFAR算法;在杂波和多干扰目标环境下,检测率分别为97.83%和98.23%。在均匀和非均匀噪声环境下,DCS-CFAR检测算法均优于ACCA-CFAR和GO-CFAR算法。结果表明,提出的DCS-CFAR检测算法在均匀和非均匀噪声环境下,均具有良好的检测性能。

In order to solve the problem that the detection performance of the radar target detector decreases badly in non-homogeneous environments.Based on the actual echo clutter distribution,a dual-censoring threshold constant false alarm rate (DCT-CFAR) detector is proposed.Dual censoring threshold is used to remove the large and small unwanted samples and real-time accurate estimate the background noise power level.Compared with the simulation and analysis results of other detectors,the proposed detector has the best detection performance and stability in multi-interfering targets,masking effect,clutter edge and other non-homogenous environments.The results show that the proposed detector still has a good detection performance in non-homogeneous environments.