SFE-SSD: 浅层特征增强的SSD算法在小目标检测问题中的研究与应用

SSD (Single Shot Multibox Dectetor)算法由于具有高速且高精度的检测性能,是目前最好的目标检测算法之一.但由于提取检测框的特征层的特征信息不足, SSD算法在小目标检测任务中表现不佳.为了解决这个问题,目前大部分方法以严重牺牲检测速度为代价提升目标检测模型的精度. 本文提出了SFE-SSD (Shallow Feature Enhancement SSD)提升SSD模型在小目标检测任务中的性能.首先我们采用反卷积操作对SSD算法中检测框金字塔特征层的最浅特征层进行特征扩张.接着通过特征融合机制对扩张后的特征层进行特征增强操作.浅层特征增强策略与SSD 的原始框提取金字塔特征层是并行结构,一定程度上是可以减少检测速度的损失.实验结果显示,我们的方法在PASCAL VOC 2007数据库上精度达到了78.4\%mAP高于SSD算法1.2\%,检测速度达到了81帧/秒,并且在小目标检测任务中有着显著的提升.     

基于相机标定的双目三维景物重建方法

介绍了一种基于相机标定的双目静止镜头下的三维景物重建方法.方法假设已经通过相机标定的方法得到两个相机的内外投影矩阵参数,通过投影矩阵和几何的方法,配合图像配准技术,恢复感兴趣场景的深度信息,达到三维景物重建的目的.本方法主要是基于几何变换的方法,巧妙避开了三维投影矩阵的计算,直接从真实图像映射为虚拟极平面图像,可以达到比较好的效果.     

Constructing the Co cyclic Structures for Crossed Copro duct Coalgebras

In this paper, we construct a cocylindrical object associated to two coal-gebras and a cotwisted map. It is shown that there exists an isomorphism between the cocyclic object of the crossed coproduct c...     

ECONOMICALLY OPTIMAL MARINE RESERVES WITHOUT SPATIAL HETEROGENEITY IN A SIMPLE TWO‐PATCH MODEL

Bioeconomic analyses of spatial fishery models have established that marine reserves can be economically optimal (i.e., maximize sustainable profit) when there is some type of spatial heterogeneity in the system. Analyses of spatially continuous models and models with more than two discrete patches have also demonstrated that marine reserves can be economically optimal even when the system is spatially homogeneous. In this note we analyze a spatially homogeneous two‐patch model and show that marine reserves can be economically optimal in this case as well. The model we study includes the possibility that fishing can damage habitat. In this model, marine reserves are necessary to maximize sustainable profit when dispersal between the patches is sufficiently high and habitat is especially vulnerable to damage.     

Semiparametric spatio‐temporal analysis of regional GDP growth with respect to renewable energy consumption levels

This article evaluates the impact of renewable energy consumption levels on GDP growth. Renewable energy sources may be significantly more costly as compared to classical (fossil) sources. Hence, high relative proportion of renewable energy has the potential of obstructing economic growth. On the other hand, in developed countries, renewable energy adoption processes are often subject to prominent state‐level incentives, administrative ingerence, and fiscal subsidies. For stratified and unbiased evaluation of the overall effect of renewables on GDP, this article applies a spatio‐temporal analysis of GDP growth at the regional NUTS2 level in a conveniently selected group of 11 spatially adjacent yet heterogeneous EU member states. As macroeconomic covariates are controlled for, along with spatial and temporal interdependencies, there is no sign of negative impact of renewable energy consumption on GDP growth. While the estimated overall effect is positive and statistically significant, its economic significance is small. Yet, given the data and economies considered, we may conclude that renewable energy consumption does not exert negative influence on economic growth rates.     

Swendsen‐Wang dynamics for general graphs in the tree uniqueness region

The Swendsen‐Wang (SW) dynamics is a popular Markov chain for sampling from the Gibbs distribution for the ferromagnetic Ising model on a graph G = (V,E). The dynamics is conjectured to converge to equilibrium in O(|V|^1/4) steps at any (inverse) temperature β, yet there are few results providing o(|V|) upper bounds. We prove fast convergence of the SW dynamics on general graphs in the tree uniqueness region. In particular, when β < β_c(d) where β_c(d) denotes the uniqueness/nonuniqueness threshold on infinite d‐regular trees, we prove that the relaxation time (i.e., the inverse spectral gap) of the SW dynamics is Θ(1) on any graph of maximum degree d ≥ 3. Our proof utilizes a monotone version of the SW dynamics which only updates isolated vertices. We establish that this variant of the SW dynamics has mixing time and relaxation time Θ(1) on any graph of maximum degree d for all β < β_c(d). Our proof technology can be applied to general monotone Markov chains, including for example the heat‐bath block dynamics, for which we obtain new tight mixing time bounds.     

Free hypermodules: a categorical approach

Abstract

In category theory, the existence of free objects is very important, especially free modules that play an important role in homological algebra. Although algebraic hyperstructures are a natural extension of algebraic structures, due to the major difference between them, study-free objects in algebraic hyperstructures become very difficult. In this article, we provide a categorical approach for the construction of free hypermodules. In fact, by considering appropriate morphisms between hypermodules, we characterize free hypermodules from three different perspectives.     

基于空间视角下产业协同集聚对区域创新的影响

选取2009-2018年中国30个省(市、自治区)的数据,运用区位商相对差异指数和熵权TOPSIS法分别测算其产业协同集聚度和区域创新水平,研究二三产业协同集聚对区域创新的影响.研究表明:产业集聚通过“知识技术、劳动力、物力”等外部规模经济增加创新产出;2009-2018年,两产业协同集聚水平由中西部向东部递减,区域创...     

Torsion simple modules over the quantum spatial ageing algebra

Various classes of simple torsion modules are classified over the quantum spatial ageing algebra (this is a Noetherian algebra of Gelfand-Kirillov dimension 4). Explicit constructions of these modules are given and for each module its annihilator is found.