| 多CPU并行结构神经网络目标识别系统 |
| |
| 引用本文： | 乔双,许廷发,宋建中. 多CPU并行结构神经网络目标识别系统[J]. 吉林大学学报(信息科学版), 2003, 0(Z1) |
| |
| 作者姓名： | 乔双 许廷发 宋建中 |
| |
| 作者单位： | 东北师范大学物理系 中国科学院长春光学精密机械与物理研究所 中国科学院长春光学精密机械与物理研究所 吉林 长春 吉林 长春 |
| |
| 摘 要： | 设计了一种用多CPU并行结构实现多目标识别的虚拟神经网络系统。该系统由图像捕获单元,3个并行工作的神经网络识别单元、运动特征计算单元、融合逻辑和相应的软件构成。实际测试表明,对18种目标的统计识别率达到90%以上,且在2.8ms之内给出识别结果,可以满足实时性要求。
|
| 关 键 词： | 并行结构 神经网络 目标识别 |
Design of parallel neural network target recognition system with many DSPs |
| |
| QIAO Shuang,XU Ting-fa,SONG Jian-zhong. Design of parallel neural network target recognition system with many DSPs[J]. Journal of Jilin University:Information Sci Ed, 2003, 0(Z1) |
| |
| Authors: | QIAO Shuang XU Ting-fa SONG Jian-zhong |
| |
| Abstract: | This paper introduces a virtual neural network system that uses many parallel CPUs to realize more targets recognition. This system is made up of image capture unit, 3 parallel neural network recognition units, mobile characteristics calculation unit, syncretism logic, and corresponding software. Practical tests indicate that the statistical recognition ratio of 18 targets is more than 90% and the result can be given in 2.8ms. So this system meets the real time demand. |
| |
| Keywords: | Parallel structure Neural network Target recognition |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
