基于Point Transformer方法的鱼类三维点云模型分类

为实现对不同鱼类的精准分类,研究共采集110尾真实鱼类的三维模型,对获取的3D模型进行基于预处理、旋转增强和下采样等操作后,获取了1650尾实验样本。然后基于Point Transformer网络和2个三维分类的对比网络进行数据集的分类训练和验证。结果表明,利用本实验的目标方法Point Transformer获得了比2个对比网络更好的分类结果,整体的分类准确率能够达到91.9%。同时对所使用的三维分类网络进行有效性评估,3个模型对于5种真实鱼类模型的分类是有意义的,其中Point Transformer的模型ROC曲线准确率最高,AUC面积最大,对于三维鱼类数据集的分类最为有效。研究提供了一种可以实现对鱼类三维模型进行精准分类的方法,为以后的智能化渔业资源监测提供一种新的技术手段。

Optimizing electronic structure through point defect engineering for enhanced electrocatalytic energy conversion

Point defect engineering endows catalysts with novel physical and chemical properties,elevating their electrocatalytic efficiency.The introduction of defects emerges as a promising strategy,effectively modifying the electronic structure of active sites.This optimization influences the adsorption energy of intermediates,thereby mitigating reaction energy barriers,altering paths,enhancing selectivity,and ultimately improving the catalytic efficiency of electrocatalysts.To elucidate the impact of defects on the electrocatalytic process,we comprehensively outline the roles of various point defects,their synthetic methodologies,and characterization techniques.Importantly,we consolidate insights into the relationship between point defects and catalytic activity for hydrogen/oxygen evolution and CO_(2)/O_(2)/N_(2) reduction reactions by integrating mechanisms from diverse reactions.This underscores the pivotal role of point defects in enhancing catalytic performance.At last,the principal challenges and prospects associated with point defects in current electrocatalysts are proposed,emphasizing their role in advancing the efficiency of electrochemical energy storage and conversion materials.

China’s Climate Policy:Mandate-Based vs.Market-Based Approaches

Mandate-based and market-based mechanisms represent two primary approaches to achieving policy objectives,yet the debate over their relative effectiveness remains unresolved.The mandate-based approach is exemplified by pilot programs for low-carbon provinces and cities,referred to as“Low-Carbon Pilot Provinces/Cities”,while the market-based mechanism is reflected in pilot programs for carbon emissions trading markets,or“Carbon Trading Pilot Programs”.This paper employs event study analysis to compare the carbon emission reduction impacts of these two approaches.Our findings reveal that the Low-Carbon Pilot Provinces/Cities achieved emissions reduction primarily by curbing economic output,without significantly reducing carbon emissions intensity.In contrast,the Carbon Trading Pilot Programs led to an increase in total carbon emissions by driving economic growth,even as they reduced carbon emissions intensity.A heterogeneity analysis further indicates that the emissions reductions observed in the Low-Carbon Pilot Provinces/Cities were predominantly concentrated in economically less-developed regions,whereas the increase in carbon emissions associated with the Carbon Trading Pilot Programs was more significant in regions with lower initial carbon emissions intensity.Against the backdrop of China’s efforts to achieve its carbon peak and neutrality goals,this paper offers valuable insights for the design of effective climate policies.

Digital Economy and Corporate Innovation-Washing--An Empirical Study on How Information Mitigates Policy Distortion

As stated in the Report to the 20th National Congress of the Communist Party of China(CPC),innovation remains at the heart of China’s modernization drive,and it is vital to optimize the allocation of innovation resources,deepen structural scientific and technological reforms,and enhance the overall performance of China’s innovation system.Government incentives have boosted firm R&D and innovation efforts;however,they have also triggered an innovation dilemma where enterprises,capitalizing on their informational advantages,resort to innovation-washing behaviors that undermine the intended purpose of the policies.Based on the information asymmetry theory,this paper conducts an empirical study on how the digital economy affects firms’innovation-washing behavior.The development of the regional digital economy could suppress firm innovation-washing behavior in the region,and such a mitigation effect is primarily caused by an increase in the number of digital industry professionals.According to our heterogeneity analysis,the digital economy has a greater impact on firm innovation-washing behavior for certain types of enterprises,including non-state-owned enterprises(non-SOEs),small and medium-sized enterprises(SMEs),enterprises in less competitive industries,and enterprises in unfavorable business environments.Our mechanism analysis revealed that the digital economy may restrain innovation-washing behavior by reducing information asymmetry between enterprises and external stakeholders.In terms of economic outcomes,the digital economy has the potential to directly influence firm innovation output while also indirectly mitigating the subsequent decline in innovation output by discouraging innovation-washing.This paper enriches the research findings on how the digital economy breaks down“information silos”and offers a potential solution to the“emphasis on input and quantity over quality and efficiency”phenomenon in science and technology innovation practices.

Deep reinforcement learning based integrated evasion and impact hierarchical intelligent policy of exo-atmospheric vehicles

Exo-atmospheric vehicles are constrained by limited maneuverability,which leads to the contradiction between evasive maneuver and precision strike.To address the problem of Integrated Evasion and Impact(IEI)decision under multi-constraint conditions,a hierarchical intelligent decision-making method based on Deep Reinforcement Learning(DRL)was proposed.First,an intelligent decision-making framework of“DRL evasion decision”+“impact prediction guidance decision”was established:it takes the impact point deviation correction ability as the constraint and the maximum miss distance as the objective,and effectively solves the problem of poor decisionmaking effect caused by the large IEI decision space.Second,to solve the sparse reward problem faced by evasion decision-making,a hierarchical decision-making method consisting of maneuver timing decision and maneuver duration decision was proposed,and the corresponding Markov Decision Process(MDP)was designed.A detailed simulation experiment was designed to analyze the advantages and computational complexity of the proposed method.Simulation results show that the proposed model has good performance and low computational resource requirement.The minimum miss distance is 21.3 m under the condition of guaranteeing the impact point accuracy,and the single decision-making time is 4.086 ms on an STM32F407 single-chip microcomputer,which has engineering application value.

Sensitivity Analysis of Structural Dynamic Behavior Based on the Sparse Polynomial Chaos Expansion and Material Point Method

This paper presents a framework for constructing surrogate models for sensitivity analysis of structural dynamics behavior.Physical models involving deformation,such as collisions,vibrations,and penetration,are devel-oped using the material point method.To reduce the computational cost of Monte Carlo simulations,response surface models are created as surrogate models for the material point system to approximate its dynamic behavior.An adaptive randomized greedy algorithm is employed to construct a sparse polynomial chaos expansion model with a fixed order,effectively balancing the accuracy and computational efficiency of the surrogate model.Based on the sparse polynomial chaos expansion,sensitivity analysis is conducted using the global finite difference and Sobol methods.Several examples of structural dynamics are provided to demonstrate the effectiveness of the proposed method in addressing structural dynamics problems.

Timely Policy Exit:Reducing Over-Investment and Driving High-Quality Firm Performance

Using data from the 11th to 14th Five-Year Plan periods(2006-2025),this study applies a Difference-in-Differences(DID)approach to assess the impact of industrial policy withdrawal.Industries that have faced policy withdrawal for over a decade are categorized as the treatment group,while consistently supported industries form the control group.The analysis examines how withdrawal affects firm total factor productivity(TFP)and investment behavior.The results show that policy withdrawal boosts firm TFP by reducing over-investment and improving the efficiency of R&D spending.This effect is particularly evident in industries with strong,competitive leading firms.Additionally,in regions with lower levels of marketization,timely policy withdrawal plays a key role in curbing over-investment.This study also highlights a dual effect of policy withdrawal:while it fosters corporate social responsibility,it may also encourage financial speculation.These findings suggest that the implementation of industrial policy should provide“timely assistance”over a limited timeframe rather than long-term support to well-established industries.As industries mature,policy support should be gradually reduced or phased out to avoid over-investment and enhance firm efficiency.

基于注意力权重PointNet++的电力走廊点云语义分割研究

传统的电力走廊点云数据的分割会出现精度低、数据局部特征捕获存在局限性等问题,为此提出了一种基于注意力权重的PointNet++网络场景分割模型。将深度学习中的PointNet++算法用于电力走廊场景分割中,再引入了空间注意力机制,帮助模型更有效地关注重要的空间区域。为此采用自制的数据集,并基于PointNet++网络模型的经典结构,在每个点集抽取(set abstraction,SA)模块中的多层感知机(multi layer perceptron,MLP)加入倒置瓶颈设计,提高对点云数据的处理效率和准确性。研究结果表明,与传统的PointNet++网络相比,改进的PointNet++网络平均交并比(mean intersection over union,mIoU)高出6.3%,加入空间注意力机制的改进模型在自制数据集上表现出更好的分割效果,尤其是在边界划分方面提升明显,验证了该方法在点云语义分割上的有效性。

位置自适应卷积PointNet++的点云数据分类方法

针对复杂场景中点云数据分类精度低问题,提出一种基于位置自适应卷积的PointNet++深度神经网络模型。由于位置自适应卷积具有较强捕捉细粒度局部特征能力,能充分获取三维点云的空间变化和几何结构特征信息,故本文在PointNet++基础上,首先通过最远点采样获取关键点,其次根据关键点使用K最近邻方法(KNN)实现分组,然后由位置自适应卷积代替原方法中的MLP提取每组的局部特征,最终完成点云分类。在2个公开的点云数据集S3DIS、Semantic3D上对本文方法进行多次对比实验,实验结果表明,本文方法在室内数据集S3DIS上的总体精度和mIoU较PointNet++网络分别提高约2.7个百分点和3.2个百分点,在室外数据集Semantic3D上的总体精度和mIoU PointNet++分别高出约2.5个百分点和2.1个百分点。

基于PointNet的钢板毛坯垛点云分割

钢板毛坯垛拆垛工序中,钢板毛坯分层厚度的估计是推钢机准确且安全执行推钢动作的关键;当前不少企业此道生产工序仍主要靠操作员人工观察的方式估算钢板厚度及与传送辊道的相对高度,其估算不准易导致碰撞,造成输送设备损坏生产中断。提出一种钢板毛坯垛智能分层方法,该方法结合现场工况环境采用激光雷达,对钢板毛坯垛进行三维点云成像,然后对采集的点云数据,用PointNet神经网络框架进行特征识别、分层分割与提取,最后对分割的不同层根据标定值,换算成真实厚度。根据现场实验结果表明,PointNet对钢板毛坯垛分割的识别率达到了87.4%,厚度估算误差低于1.2cm,根据钢厂钢板毛坯规格表(3种规格150、160、180cm)可以准确估计出钢板毛坯厚度的规格,识别速度15帧/s,满足现场工况要求。

基于轻量化PointNet网络的林果园喷雾作业靶标实时识别方法

为了进一步提高喷雾机器人靶标检测的精准性、实时性和应用部署的实用性,该研究提出一种基于轻量化PointNet网络的林果园喷雾作业靶标实时识别方法。首先通过区域提取降采样、地面分割和改进DBSCAN聚类等点云预处理方法提取原始点云中的靶标;然后通过移动最小二乘上采样将靶标点云转化为满足点云识别网络输入要求的点云数据;最终通过在PointNet网络中引入残差模块和改进循环剪枝算法轻量化PointNet网络,完成林果树靶标的实时识别。试验结果表明,在ModelNet40数据集上,轻量化PointNet网络可达89.7%的准确率;在实际苗圃环境的试验中,该研究方法对靶标的识别准确率可达92.49%,同时误识率与拒识率分别为13.4%和6.47%,相较PointNet网络识别准确率提升了4.38个百分点,误识率和拒识率分别降低了7.2和4.07个百分点;轻量化PointNet网络识别准确率仅比PointNet++网络低1.14个百分点,误识率和拒识率分别高了0.9和1.12个百分点。但是轻量化PointNet网络的模型参数量较PointNet网络和PointNet++网络的模型参数量显著减少,仅为PointNet网络的11.5%,PointNet++网络的27.02%;运算量相较PointNet网络、PointNet++网络分别减少13.3和76.79个百分点。该研究提出的轻量化PointNet网络具有较高的实时性、精确性和鲁棒性,能够满足林果园喷雾作业的靶标识别需求,可为林果园喷雾作业靶标实时识别提供参考。

基于K F-PointNet++的油菜植株点云分割算法

针对传统的点云分割算法精度低、鲁棒性差的问题,提出了基于K近邻算法和特征融合的深度点云网络(point clouds network++based on K-nearest neighbor algorithm and feature fusion,K F-PointNet++)三维点云分割算法。该算法首先采用了K近邻(K-nearest neighbors,K NN)算法对点云进行分组;其次,将点云网络(point clouds network,PointNet)中的局部特征与中心点全局特征进行拼接,增强算法对几何细节和全局上下文信息的捕捉能力,从而提高算法的分割精度和鲁棒性,实现了对油菜点云器官的精准分割。使用自制的油菜点云数据集进行实验,结果表明,K F-PointNet++算法在油菜点云分割中的总体精度(overall accuracy,OA)可达97.1%,平均交并比(mean intersection over union,mIoU)为86.4%。该算法在分割性能方面明显优于PointNet、深度点云网络(PointNet++)和核点卷积(kernal point convolution,KPConv),可以为油菜表型研究提供可靠基础。

基于PointNet++的邻域特征增强点云语义分割方法

随着智能驾驶、机器人导航等以点云为基础的应用蓬勃发展,点云语义分割逐渐成为研究热点。然而,现有的点云语义分割方法存在局部特征提取不充分、特征融合不完整的缺陷。针对这些不足,提出了对应的解决方案。对于局部特征提取不充分的现象,通过嵌入邻域点的坐标、方向、距离等相关信息去关联邻域点的显式特征。对于特征融合不完整的现象,提出了一种最大池化与自注意力池化相结合的混合池化方法。网络架构基于PointNet++,并结合提出的局部特征提取和融合方法,在S3DIS数据集上的实验结果表明,与基线方法PointNet++相比,各评价指标都有不同程度的提高,证实了新方法的有效性和优越性。

The impacts of agricultural and rural economic policy system on agricultural non-point source pollution in China

Agricultural and rural economic policy system is one main driving force for the evolvement of agricultural Non-Point Source （NPS） pollution. In this paper, the main policies that influence agricultural NPS pollution are chosen, and a method to evaluate the impacts of agricultural and rural economic policy system on agricultural NPS pollution is brought forward. According to this, the questions about how and to what degree the policy system influence on agricultural NPS pollution are discussed.

基于Point Transformer v2的点云枝叶分离方法研究

准确高效的点云枝叶分离对精确计算森林树木的垂直参数至关重要。然而,当前的研究方法计算成本高,且依赖先验知识导致泛化能力不足。针对以上问题,文章提出利用基于点特征的Transformer网络进行自动化的森林场景三维点云的枝叶分离研究。该方法使用Point Transformer v2网络,首先利用网格编码模块提取可学习的局部结构关系,保留点云的几何拓扑结构;其次使用分组注意力实现多通道联合学习,降低特征的冗余度,提高计算的效率;最后构建了基于点的Transformer网络实现高精度森林树木三维点云语义分割,降低了对于先验知识的需求。使用地基激光扫描仪获取的加拿大和芬兰7个不同树种样地的三维点云数据,进行枝叶分离实验和精度评价。实验结果表明,网络整体精度(OA)为94.42%,mIoU为78.89%,能够适应不同树种、不同点云密度的森林场景的枝叶分离。

基于改进PointNet++的船体分段合拢面构件智能识别算法研究

[目的]三维扫描仪获得的船体分段合拢面点云数据,具有精度高、数据量大的优势,能够很好地反映分段合拢面的建造状况。由于现有的PointNet++网络无法处理大容量点云数据,因此提出一种基于改进PointNet++的船体分段合拢面构件智能识别算法,实现针对大容量船体分段合拢面点云数据构件的智能识别。[方法]基于超体素生长理论对船体分段合拢面点云数据进行分割及简化,构建船体分段合拢面点云数据集,并使用该数据集训练基于深度学习理论改进的PointNet++网络。[结果]网络模型在船体分段合拢面点云数据训练集和测试集上的收敛结果趋于稳定,在测试集上识别准确率达到90.012%。[结论]该方法具有良好的识别能力,能够完成船体分段合拢面构件的智能识别。

基于PointNet++进行附属设施语义分割的隧道收敛变形分析

随着城市轨道交通日趋广泛,隧道结构变形引起的地铁安全事故凸显,亟需对运营期隧道进行变形检测。隧道衬砌作为隧道变形分析的研究对象,衬砌内表面存在的大量附属设施影响隧道收敛变形分析精度。为了提高变形分析精度,解决点云处理环节中存在的自动化程度低的问题,提出基于PointNet++点云语义分割的隧道收敛变形分析方法。首先利用深度学习方法进行点云语义分割,对隧道衬砌附属设施进行自动滤除。然后对隧道衬砌进行断面提取,利用随机抽样一致性算法(Random Sample Consensus,RANSAC)对隧道断面点云进行采样,分析隧道收敛变形程度,从Z+F PROFILER 9012A激光断面扫描仪获取山东省济南市地铁盾构隧道点云实测数据上并进行应用。研究结果表明:所提出的处理方法可以有效地将大规模隧道衬砌与连接紧密的附属设施分离出来,隧道附属设施总体分类精度达到96%,滤波结果较好地保留了隧道衬砌原始形态特征。在对隧道整体和局部收敛变形分析的重复性验证中,测试区间内隧道整体变形精度往返测长半轴平均偏差为1.04 mm,短半轴平均偏差为0.9 mm,测试区间内隧道局部收敛变形往返测标准差最小为0.773 mm,最大为0.938 mm,可以满足隧道收敛变形分析的精度需求。研究结果可以有效提升处理大规模隧道数据的自动化程度,具有良好的有效性与可靠性,对运营期地铁隧道收敛变形检测或监测有较好的实践应用意义。

Diagnosis of indirectly driven double shell targets with point-projection hard x-ray radiography

We present an application of short-pulse laser-generated hard x rays for the diagnosis of indirectly driven double shell targets. Coneinserted double shell targets were imploded through an indirect drive approach on the upgraded SG-II laser facility. Then, based on thepoint-projection hard x-ray radiography technique, time-resolved radiography of the double shell targets, including that of their near-peakcompression, were obtained. The backlighter source was created by the interactions of a high-intensity short pulsed laser with a metalmicrowire target. Images of the target near peak compression were obtained with an Au microwire. In addition, radiation hydrodynamicsimulations were performed, and the target evolution obtained agrees well with the experimental results. Using the radiographic images, arealdensities of the targets were evaluated.

Point Cloud Classification Using Content-Based Transformer via Clustering in Feature Space

Recently, there have been some attempts of Transformer in 3D point cloud classification. In order to reduce computations, most existing methods focus on local spatial attention,but ignore their content and fail to establish relationships between distant but relevant points. To overcome the limitation of local spatial attention, we propose a point content-based Transformer architecture, called PointConT for short. It exploits the locality of points in the feature space(content-based), which clusters the sampled points with similar features into the same class and computes the self-attention within each class, thus enabling an effective trade-off between capturing long-range dependencies and computational complexity. We further introduce an inception feature aggregator for point cloud classification, which uses parallel structures to aggregate high-frequency and low-frequency information in each branch separately. Extensive experiments show that our PointConT model achieves a remarkable performance on point cloud shape classification. Especially, our method exhibits 90.3% Top-1 accuracy on the hardest setting of ScanObjectN N. Source code of this paper is available at https://github.com/yahuiliu99/PointC onT.