As an important computer vision task, 3D human pose estimation has received widespread attention and many applications have been derived from it. Most previous methods address this task by using a 3D pictorial structure model which is inefficient due to the huge state space. We propose a novel approach to solve this problem. Our key idea is to learn confidence weights of each joint from the input image through a simple neutral network. We also extract the confidence matrix of heatmaps which reflects its feature quality in order to enhance the feature quality in occluded views. Our approach is end-to-end differentiable which can improve the efficiency and robustness. We evaluate the approach on two public datasets including Human3.6M and Occlusion-Person which achieves significant performance gains compare with the state-of-the-art.
In the widely used field of panoramic image stitching, the key technologies mainly cover two parts, i.e., image registration and image fusion. In order to achieve low-cost and real-time processing, researchers often design dedicated circuits for various image stitching algorithms. Many studies focus on image registration algorithms and ignore image fusion algorithms, let alone dedicated circuit design. In addition, to reduce the ghosting effect and deformation caused by seams in stitching, finding the best seam line based on the overlapping area of the stitching image is crucial, which directly affects the quality of image stitching and fusion. In order to solve the above problems and achieve the efficient fusion of registered images, an image stitching and fusion algorithm circuit based on a dynamic programming algorithm to search for seam lines was proposed. Comprehensive experimental results and a theoretical analysis based on Cyclone IV FPGA devices show that, with a clock frequency of 100 MHz, the proposed circuit takes about 7.04 ms to carry out the fusion processing of two 486 × 643 images, and the corresponding frame rate is approximately 142 FPS, achieving a perfect real-time stitching effect and meeting the demand for real-time image processing. After a theoretical derivation and comparison with other similar works, its processing speed is better than four state-of-the-art implementations.
The operation of the Insulated Gate Bipolar Transistor (IGBT) is closely related to the operating state of the drive circuit. Based on the theory of analog circuit fault diagnosis, this paper proposed an incipient fault diagnosis method combing Empirical Wavelet Transform (EWT) and Residual Network (ResNet) for IGBT drive circuit which has an EXB-841 driver module as the core. Firstly, the drive circuit was divided into the driving function (DF) part and the short-circuit protection (SP) part according to its working principle and basic structure. And the sensitivity analysis was performed on all the main components in both sections to select the test objects. Secondly, the circuit signals were obtained by fault injection and Monte Carlo analysis and then decomposed by improved EWT based on Scale Space (SS) to construct the datasets. Finally, based on the structure of ResNet18, 1D-ResNet was established and trained on the collected datasets to achieve deep feature extraction and fault classification. Simulation results show that the incipient fault diagnosis accuracy of DF part and SP part is 99.55% and 97.55% respectively.
随着集成电路制程趋于极限,登纳德缩放定律逐步失效,芯片的功率密度逐渐提升,尤其是在5G、物联网以及高性能计算快速发展的驱动下,单芯片面积也在增大,热耗散问题日趋严重,传统的冷却方式已无法保证芯片的可靠工作.将热沉制备在芯片内部可以避免封装材料的导热热阻和多层界面热阻,提升冷却性能和冷却效率.学术界针对芯片的嵌入式微流体冷却开展了大量卓有成效的研究和探索,不断提出新型通道结构设计方案,包括平行长直通道、歧管通道、射流通道等.旨在于优化泵功和热阻,在小压降下实现高效冷却.然而,随着芯片面积的增大,在限域空间实现高效冷却将更加困难,工艺难度和制造成本限制了嵌入式液冷的大规模商业化使用,目前在实际IC芯片内演示的冷却方案验证了嵌入式冷却的性能,但复杂度高,兼容性差,冷却性能有待进一步提升.尤其是在3D封装架构下,需要提出兼容小型化、高密度封装的通道结构,通过协同设计,在保证电学互连的前提下实现层间冷却.在优化通道结构设计的同时,还需要简化工艺,降低成本,提升嵌入式微流体冷却的工艺可靠性和长期工作可靠性,才能推进嵌入式微流体冷却技术的实际应用.
随机森林具有训练速度快、不容易过拟合、易于实现等优点成功应用于各种领域.针对在芯片设计后仿阶段需要对不同存储存储单元大小、电压、温度等情况进行功耗测试,并且测试一次的时间很长的问题,提出了一种基于麻雀搜索算法(SSA)与随机森林(RF)相结合的功耗预测方法.首先,对14nmSRAM后仿的单元库进行表征,找出合适的特征变量,得到特征数据后构建训练测试集;然后对特征变量进行特征重要性分析,按照特征重要性排序;最后使用随机森林模型进行回归预测,并引入了麻雀搜索算法以寻找出均方根误差最小时的模型参数.与线性回归模型、支持向量回归模型等相比,SSA-RF收敛精度高并且训练速度快,SSA-RF模型的R2值为0.97左右.此外,在数据量较少的情况下其R2的值也能达到0.95左右,构建了一个较好的预测模型,为减少功耗测试时间提供了一种可行的方案,可以为设计人员留下更多的时间去优化电路.
