To obtain high-resolution of the subsurface structure, we modeled multidepth slanted airgun sources to attenuate the source ghost. By firing the guns in sequence according to their relative depths, such a source can b...To obtain high-resolution of the subsurface structure, we modeled multidepth slanted airgun sources to attenuate the source ghost. By firing the guns in sequence according to their relative depths, such a source can build constructive primaries and destructive ghosts. To evaluate the attenuation of ghosts, the normalized squared error of the spectrum of the actual vs the expected signature is computed. We used a typical 680 cu.in airgun string and found via simulations that a depth interval of 1 or 1.5 m between airguns is optimum when considering deghosting performance and operational feasibility. When more subarrays are combined, preliminary simulations are necessary to determine the optimum depth combination. The frequency notches introduced by the excess use of subarrays may negatively affect the deghosting performance. Two or three slanted subarrays can be combined to remove the ghost effect. The sequence combination may partly affect deghosting but this can be eliminated by matched filtering. Directivity comparison shows that a multi-depth slanted source can significantly attenuate the notches and widen the energy transmission stability area.展开更多
为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度...为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度特征提取模块,获取到具有全局上下文信息和位置信息的多尺度特征图像。通过残差学习网络对深度图进行优化,防止多次卷积操作出现重建边缘模糊的问题。通过分类的思想构建focal loss函数增强网络模型的判断能力。由实验结果可知,该算法在DTU(technical university of denmark)数据集上和CasMVSNet(Cascade MVSNet)算法相比,在整体精度误差、运行时间、显存资源占用上分别降低了14.08%、72.15%、4.62%。在Tanks and Temples数据集整体评价指标Mean上该模型优于其他算法,证明提出的基于自适应空间特征增强的多视图深度估计算法的有效性。展开更多
为提高多无人船编队系统的导航能力,提出了一种基于注意力机制的多智能体深度确定性策略梯度(ATMADDPG:Attention Mechanism based Multi-Agent Deep Deterministic Policy Gradient)算法。该算法在训练阶段,通过大量试验训练出最佳策略...为提高多无人船编队系统的导航能力,提出了一种基于注意力机制的多智能体深度确定性策略梯度(ATMADDPG:Attention Mechanism based Multi-Agent Deep Deterministic Policy Gradient)算法。该算法在训练阶段,通过大量试验训练出最佳策略,并在实验阶段直接使用训练出的最佳策略得到最佳编队路径。仿真实验将4艘相同的“百川号”无人船作为实验对象。实验结果表明,基于ATMADDPG算法的队形保持策略能实现稳定的多无人船编队导航,并在一定程度上满足队形保持的要求。相较于多智能体深度确定性策略梯度(MADDPG:Multi-Agent Depth Deterministic Policy Gradient)算法,所提出的ATMADDPG算法在收敛速度、队形保持能力和对环境变化的适应性等方面表现出更优越的性能,综合导航效率可提高约80%,具有较大的应用潜力。展开更多
针对多视图三维重建中存在的内存和时间消耗过大、高分辨率重建完整性差等问题,提出一种基于深度学习的多视图重建网络。网络由特征提取模块、级联的Patchmatch模块和深度图优化模块组成。首先,设计U型的特征提取模块,提取多阶段特征图...针对多视图三维重建中存在的内存和时间消耗过大、高分辨率重建完整性差等问题,提出一种基于深度学习的多视图重建网络。网络由特征提取模块、级联的Patchmatch模块和深度图优化模块组成。首先,设计U型的特征提取模块,提取多阶段特征图,并在每个阶段引入相对位置编码的局部自注意力层,捕捉图像中的局部细节和全局上下文,提升网络特征提取性能。其次,设计深度残差网络,通过密集连接和残差结构对特征进行融合,充分利用彩色图像先验知识来约束深度图,提升深度估计的准确性。在公开数据集DTU(Technical University of Denmark)上进行测试,实验结果表明,三维重建质量到了有效的提升,与PatchmatchNet相比在完整性上提升了6.1%,在整体性上提升了2.5%,与其他的SOTA(State-Of-The-Art)方法相比,在完整性和整体性上都得到了较大提升。展开更多
基金financially supported by the national 863 program(2013AA064202)Marine subject interdisciplinary and guidance fund of Zhejiang University(188040+193414Y01)
文摘To obtain high-resolution of the subsurface structure, we modeled multidepth slanted airgun sources to attenuate the source ghost. By firing the guns in sequence according to their relative depths, such a source can build constructive primaries and destructive ghosts. To evaluate the attenuation of ghosts, the normalized squared error of the spectrum of the actual vs the expected signature is computed. We used a typical 680 cu.in airgun string and found via simulations that a depth interval of 1 or 1.5 m between airguns is optimum when considering deghosting performance and operational feasibility. When more subarrays are combined, preliminary simulations are necessary to determine the optimum depth combination. The frequency notches introduced by the excess use of subarrays may negatively affect the deghosting performance. Two or three slanted subarrays can be combined to remove the ghost effect. The sequence combination may partly affect deghosting but this can be eliminated by matched filtering. Directivity comparison shows that a multi-depth slanted source can significantly attenuate the notches and widen the energy transmission stability area.
文摘为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度特征提取模块,获取到具有全局上下文信息和位置信息的多尺度特征图像。通过残差学习网络对深度图进行优化,防止多次卷积操作出现重建边缘模糊的问题。通过分类的思想构建focal loss函数增强网络模型的判断能力。由实验结果可知,该算法在DTU(technical university of denmark)数据集上和CasMVSNet(Cascade MVSNet)算法相比,在整体精度误差、运行时间、显存资源占用上分别降低了14.08%、72.15%、4.62%。在Tanks and Temples数据集整体评价指标Mean上该模型优于其他算法,证明提出的基于自适应空间特征增强的多视图深度估计算法的有效性。
文摘针对多视图三维重建中存在的内存和时间消耗过大、高分辨率重建完整性差等问题,提出一种基于深度学习的多视图重建网络。网络由特征提取模块、级联的Patchmatch模块和深度图优化模块组成。首先,设计U型的特征提取模块,提取多阶段特征图,并在每个阶段引入相对位置编码的局部自注意力层,捕捉图像中的局部细节和全局上下文,提升网络特征提取性能。其次,设计深度残差网络,通过密集连接和残差结构对特征进行融合,充分利用彩色图像先验知识来约束深度图,提升深度估计的准确性。在公开数据集DTU(Technical University of Denmark)上进行测试,实验结果表明,三维重建质量到了有效的提升,与PatchmatchNet相比在完整性上提升了6.1%,在整体性上提升了2.5%,与其他的SOTA(State-Of-The-Art)方法相比,在完整性和整体性上都得到了较大提升。
