通道空间深度感知的轻量化水下目标检测

提出了一种通道空间深度感知的轻量化水下目标检测网络CSDP-L-YOLO;该网络基于YOLOv5网络进行改进,由特征感知模块和双注意门控策略组成;特征感知模块旨在将解码器中的多级特征自适应抑制或增强,优化类内学习的一致性,解决水下场景复杂导致的误检和漏检问题;通过线性操作和混洗结构生成特征映射,减少冗余特征的融合和计算,以减少模型的参数量和计算量;双注意门控策略是在编码器中同时引入并发通道空间挤压-激励机制模块和卷积注意力模块,进一步关注强相关性特征,增强模型对特征的敏感度;实验结果表明,与基线模型YOLOv5-s相比,mAP提高了2.4%,节省了20%参数量和15.8%计算量,检测速度提升了8.2 ms;此外,与目前较为先进的YOLOv8模型相比,mAP提高了1.9%。

基于梯度指导的双通道深度压缩感知图像重建方法

针对压缩感知重建图像边界部分模糊和畸变的问题,将图像的先验特性与深度学习网络相结合,提出了一种基于梯度指导的双通道深度压缩感知图像重建方法。该方法利用梯度图像和原始图像从边缘和纹理2个方面构建了双通道的深度网络模型。一方面,通过梯度分支来恢复高质量梯度图,为最后的重建图像提供额外的结构先验;另一方面,提出了梯度损失,图像的梯度约束有助于重建网络更专注于几何结构。在原图通道中采用混合卷积残差密集连接模块,扩大感受野的同时提取丰富的细节信息。实验结果表明,本方法与其他方法相比,取得了更高的重建质量,特别是在图像边界部分的恢复上有显著提升。

基于转置注意力的多尺度深度融合单目深度估计

单目深度估计是计算机视觉领域中一项基础任务,其目标是通过单张图像预测深度图,并获取每个像素位置的深度信息。本文提出一种新的单目深度估计网络结构,旨在进一步提高网络的预测准确性。转置注意力机制在降低参数量和计算量的同时引入了自注意力机制,以关注图像中的特定区域,并结合不同通道之间的信息。这种机制能够有效地关注到图像中的细小区域和边缘信息,并进行学习。本文还提出一种改进的转置注意力机制,以更少的参数量保留语义信息。多尺度深度融合根据不同通道提取不同深度特征的特点,计算每个通道的平均深度,以增强模型的深度感知能力。此外,它能够建模垂直距离的长距离关系,有效地分离物体之间的边缘,有助于减少细粒度信息的损失。最后,本文在NYU Depth V2数据集和KITTI数据集上进行实验,验证了所提出模块的有效性,并取得了出色的性能表现。

基于改进深度学习的航拍滑坡检测方法

为及时发现滑坡险情展开应急救援,提出一种结合逐通道不同阈值的深度残差收缩网络(DRSN-CW)方法的更快速区域卷积神经网络(Faster-RCNN)模型的航拍图像滑坡检测算法。利用图像增强的伽马变换、高斯滤波方法提高图片的质量;使用群组归一化方法消除batchsize大小对模型的影响;为减少噪声以及无关特征区域的干扰,采用DRSN-CW网络对滑坡检测目标进行精准定位。实验结果表明,改进后模型与先前模型相比F1值以及平均精度分别增加了10.7%、10.2%,可以有效检测滑坡险情。

基于深度学习卷积神经网络的人体行为识别研究

基于深度学习卷积神经网络人体行为识别是深度学习研究中一个重要的研究方向.文章首先介绍了深度学习在人体行为识别的应用现状以及两大重要的人体行为识别数据集:Weizmann数据集和KTH数据集,然后概述了深度学习特征提取方法,最后重点综述了双通道深度卷积神经网络的原理、结构和具体设计,以及其重要意义和优势.

使用mentalray渲染打造游戏场景

文章主要讲解了怎样在Maya制作一个游戏场景的过程。

基于DTL与DC-ResNet的非侵入式多标签负荷识别方法

针对目前负荷识别中存在的人工挖掘特征困难的问题,同时为了进一步提升识别精度,提出了一种基于深度转换学习(DTL)与双通道深度残差神经网络(DC-ResNet)的非侵入式多标签负荷识别方法。首先运用改进的DTL从原始的负荷数据中自动提取有效的负荷特征,可以达到降维的效果。然后将负荷数据按时间滑窗生成特征图作为DC-ResNet的输入,利用卷积核自动提取时间尺度上的有效特征。DC-ResNet的一条通道用于提取大类特征预测负荷的开关状态,另一条通道用于提取小类特征预测负荷的运行模式,综合两条通道的输出可以得到最后的多标签分类结果。实验结果表明,经DTL提取后的负荷数据更易被识别,DC-ResNet在总体和单一设备的识别上都具有更好的效果,而且模型所需的训练时间较短。

Experimental study of entrainment behavior of debris flow over channel inflexion points

On-spot observation and field reconnaissance of debris flows have revealed that inflexion points in the longitudinal profile of a movable channel may easily become unstable points that significantly affect their entrainment behavior.In this study,small-scale flume experiments were performed to investigate the entrainment characteristics of debris flows over two types of inflexion points,namely,a convex point,which has an upslope gradient that is less than the downslope gradient,and a concave point,which has an upslope gradient that is greater than the downslope gradient.It was observed that when debris flowed over a convex point,the entrainment developed gradually and progressively from the convex point in the downstream direction,and the primary control factors were the slope gradient and friction angle.Conversely,when debris flowed over a concave point,the entrainment was characterized by impacting and impinging erosion rather than traditional hydraulic erosion,and the impingement angle of the flow significantly determined the maximum erosion depth and outflow exit angle.An empirical relationship between the topography change and the control factors was obtained from the experimental data.

Geometry and Dynamic Forces along an Inclined Hydraulic Jump over a Rough Channel Bed

In this investigation, based on previous measurements, the geometry and dynamic forces along an inclined （angle φ, with 0^0 〈 qφ φ 8^0 ） hydraulic jump over a variation of rough and smooth channel beds, are examined and compared among them. The roughness is produced through transverse strips （of square cross section） covering the entire channel width and at various normal distributions along the channel. Froude numbers are varying up to 18.5 and conjugate depths＇ ratios up to 18.4. The water free surface profiles, the conjugate depths＇ ratios, the jumps＇ dimensionless lengths are examined and a prediction of exercised forces, in dimensionless terms, is also presented. The results and especially the latter force prediction may help the hydraulic and structural engineers when dealing with such hydraulic jumps.

Application of Index Based on the Land Surface Temperature to Estimate the Threshold Wind Speed for Saltation Activity

Land-surface conditions, such as surface roughness and SWC （soil-water content）, control the saltation activity and dust emission in northeast Asia. Information on spatial and temporal changes in surface SWC is needed for dust-modeling systems used to predict dust events with the aim of preventing the damage they cause. A MTVDI （modified temperature-vegetation dryness index） was tested to see if it could reproduce the surface SWC observed in Zhangye, China, and the Tottori Sand Dunes of Japan, and the threshold wind speed at the Tottori Sand Dunes. MTVDI was calculated from land-surface temperature using the MODIS （moderate resolution imaging spectroradiometer） product, and the aerodynamic minimum and maximum surface temperatures were estimated based on meteorological data. A greater correlation is seen between MTVDI and SWC than between SWC from AMSR-E （advanced microwave scanning radiometer-earth observing system） and SWC in Zhangye. The threshold wind speed for saltation activity decreased with increasing MTVDI, that is, with drying of the soil surface of the Tottori Sand Dunes. The correlation between MTVDI and threshold wind speed is statistically significant （R2 = 0.2987）.

Design and calibration of a new high-definition three-dimensional laparoscopic system

We present a high-definition （HD） 3D laparoscopic system including a dual channel optical system, two cameras, a camera control unit （CCU）, and an HD 3D monitor. This laparoscopic system is capable of outputting dual high-definition videos and providing vivid 3D images. A modified pinhole camera model is used for camera calibration and a new method of depth measurement to improve precision. The average error of depth measurement measured by experiment （about 1.13 mm） was small in proportion to the large range in distance of the system （10-150 mm）. The new method is applicable to any calibrated binocular vision system.