用于小目标检测的TDNLMS自适应预测器结构优化

TDNLMS Adaptive Filter Architecture Optimization for Small Objects Detection

讨论用于小目标检测的 TDNL MS( Two- Dim ensional Normalized L east Mean Square)自适应预测器 VL SI实现中的结构优化问题 .通过分析小目标图像的特点 ,使预测器直接处理隔行扫描图像 ,从而取消了帧存环节 ,大大减少了片内存储器数量 .另外 ,在基本保持检测性能的基础上 ,通过对算法进行优化 ,提高了预测器的工作速度 ,达到实时图像处理 .实验证明本方案不仅提高了预测器的数据通过率 ,而且简化了硬件电路 ,降低了设计难度 ,是一种为合理的方案 .

The architecture optimization of TDNLMS adaptive filter's VLSI implementation is discussed here. TDNLMS adaptive filter can suppress strong clutter in image data, and enhance SCR of small objects. By analyzing the feature of small object image, we cancel large volume on chip memory, and replace division by shift operation, so that the adaptive filter can processing image data in real time. These two methods can reduce adaptive filter latency and simplify the VLSI implementation. Experiment results show that performance of adaptive filter after architecture optimization for small object detection is not decreased significantly.