基于多DSP混合结构的Gabor小波神经网络图像目标识别

提出了一种基于多DSP混合结构的Gabor小波神经网络图像目标识别新方法.利用TMS320C5409设计了多DSP混合结构系统,根据并-串结构系统的特点,设计了Gabor小波神经网络算法.算法被分成不同的并-串结构进行运算,利用串行的DSP-1进行Gabor小波变换提取图像目标的特征向量,并输入到采用不同网络结构的并行多DSP进行BP网络运算,串行的DSP6对BP网络输出的后验概率进行加权平均,给出分类结果.对9种飞机目标进行了分类识别仿真实验.实验结果表明,该方法应用于多目标识别时,识别时间为2.8 ms,识别率达到98%.

Gabor Wavelet Neural Network Image Target Recognition Base on Multi-DSP Combined Structures

A novel method of Gabor wavelet neural network image target recognition is presented based on multi-DSP combined structures.Multi-DSP combined structure systems have been designed by TMS320C5409 according to the character of parallel-series system,and the algorithm of Gabor wavelet neural network is presented.The algorithms are divided into different parallel-series structures in operation,the feature vectors of the image target are extracted by series DSP1,and input in parallel many DSP using different net structures to operate the BP algorithms, the classified result was given,and adding averaged to BP net post-probability by series DSP6.the characteristic of the neural networks parallel algorithms was exerted.The simulink experiment was done by the method applied on nine types of plane target recognition,the results indicated that the recognition time was 2.8 ms,and the recognition rate 98%.