In order to meet the demands of real-time statistics for classification, this paper proposes a design for real-time intelligent recognition system of plankton target with high recognition rate. This paper acquires real-time image with CCD sensor, develops target frame selection in the processing of frames. A fast segmentation based on background subtraction instead of traditional iteration is proposed. Fast linear classifier with adaptive feature extraction is achieved for target classification. At the same time, this paper achieves parallel processing with load balancing. Experimental results show that the system reaches real-time recognition while average single-frame recognition time is less than 29.1 ms and recognition rate is as high as 91%.%根据浮游生物目标的现场分类统计需求,提出一种改进的实时目标识别系统设计方案.使用CCD传感器进行实时图像采集,在逐帧图像处理中引入目标帧筛选以减少运算.采用背景减算法替代传统的迭代算法实现快速目标分割,利用自适应特征选取的快速线形分类器进行目标分类,并通过多核负载均衡实现并行处理.实验结果表明,该系统的平均单帧识别时间低于29.1 ms,识别率高达91％,达到实时识别要求.
展开▼
