排序方式: 共有13条查询结果,搜索用时 0 毫秒
1.
人工智能在机器人控制中得到广泛应用,机器人控制算法也逐渐从模型驱动转变为数据驱动。深度强化学习算法可在复杂环境中感知并决策,能够解决高维度和连续状态空间下的机械臂控制问题。然而,目前深度强化学习中数据驱动的训练过程非常依赖计算机GPU算力,且训练时间成本较大。提出基于深度强化学习的先简化模型(2D模型)再复杂模型(3D模型)的机械臂控制快速训练方法。采用深度确定性策略梯度算法代替机械臂传统控制算法中的逆运动学解算方法,直接通过数据驱动的训练过程控制机械臂末端到达目标位置,从而减小训练时间成本。同时,对于状态向量和奖励函数形式,使用不同的设置方式。将最终训练得到的算法模型在真实机械臂上进行实现和验证,结果表明,其控制效果达到了分拣物品的应用要求,相比于直接在3D模型中的训练,能够缩短近52%的平均训练时长。 相似文献
2.
针对基于深度卷积神经网络的 RGB-D 显著性检测性能差等问题, 提出利用注意机制和多尺度跨模态融合进行 RGB-D 显著性检测的方法. 首先采用多尺度残差注意模块对骨干网络提取的特征进行预处理; 然后提出多尺度跨模态融合策略, 对高层 RGB 特征和深度特征进行融合, 获得初始显著图; 最后采用边界细化模块细化初始显著图中目标的边界, 使最终显著图包含敏锐的边界和完整的突出目标. 在 5 个基准数据集上与 10 种先进方法进行实验的结果表明, 所提方法在 4 个评价指标上均处于前 3 名; 尤其是在 NJUD 和 SIP 数据集上, 该方法在 4 个指标上提升了0.5%~1.5%. 相似文献
3.
4.
To provide a certain level of Quality of Service (QoS) guarantees for multiuser wire-less downlink video streaming transmissions, we propose a multiuser scheduling scheme for QoS guarantees. It is based on the classic Queue-Length-Based (QLB)-rate maximum scheduling algorithm and integrated with the delay constraint and the packet priority drop. We use the large deviation principle and the effective capacity theory to construct a new analysis model to find each user’s queue leng-th threshold (delay constraint) violation prob-ability. This probability corresponds to the upper bound of the packet drop probability, which indicates a certain level of statistical QoS guarantees. Then, we utilize the priority information of video packets and introduce the packet priority drop to further improve the quality perceived by each user. The simu-lation results show that the average Peak Signal to Noise Ratio (PSNR) value of the priority drop is 0.8 higher than that of the non-priority drop and the PSNR value of the most badly damaged video frame in the priority drop is on an average 4 higher than that of the non-priority drop. 相似文献
5.
灵活宏块排序FMO(Flexible Macroblock Ordering)是H.264/AVC标准提出的一项非常有效的容错机制,其中包括六种标准的FMO方式,其针对不同的视频序列以及信道特点各有优劣。通过仿真分析了标准中六种FMO方式的特点,提出了一种新的双螺旋FMO方式。仿真结果表明该方式在一定的编码码率条件下能取得更好的信道传输适应性,且容错效果优于大多数标准的FMO方式。 相似文献
7.
8.
9.
该远距离高精度温度测量系统由数据拟合模块、红外测温模块、超声波测距模块、数据处理模块以及按键显示模块组成。系统通过数据拟合模块实现采集的大量“输入—目标”样本向量数据对人工神经网络(ANN)进行训练,并三阶拟合出修正函数。实际测量时,数据处理模块采用已获得的修正函数对超声波测距模块测量的距离系数d与红外测温模块测量的初步温度进行融合运算,最终输出被测单位修正后的温度。 相似文献
10.
With relatively high transmission capacity and usually unconstrained connections, IEEE802.11 WLANs provide the ideal infrastructure for pervasive video content sharing and communications. However, the delivery of high-performance video streams over 802.11 WLANs remains a challenging task due to the inherent characteristics of compressed video and dynamic channels. In this paper, we present a brief survey of various recent innovations that have been developed to enhance the Quality of Service (QoS) performance for video over WLANs. Based on the application scenarios, the solutions have focused mainly on three network layers, that is, Application layer (APP), Media Access Control layer (MAC), and Physical layer (PHY). After reviewing the video compression technology, we first examine various single-layer solutions for video over WLANs. We then discuss several cross-layer solutions that take advantage of mutual interactions between different network layers. Finally, several technical issues beyond QoS performance, including energy and security, are also addressed. We conclude that the application of video over WLANs will continue to increase in future. 相似文献