介绍了三种类型的光电混合神经网络系统,重点研究了衍射引起的误差对系统输出的影响。根据衍射、光学信息处理和神经网络理论,采用实验中的参量和输入数据,对衍射造成的输出误差做了仿真分析。分析表明,近场衍射造成较大的输出误差。近场与远场衍射综合作用时,误差因输入图像模式不同而差异较大;其中,当输入较小的简单图像时相对误差较大;而对实验中实际采用的复杂图像,相对误差较小。利用线性回归方法对输出数据做了校正,并分析了其可行性,校正后的数据误差降低一个量级。衍射误差对实验中识别率的影响可以控制在较小的范围内,识别率可以保持在97.7%以上。 Three kinds of optoelectronic hybrid neural networks are introduced, focusing on the influence of diffraction error on the output of the system. According to the theory of diffraction, optical information processing and neural network, the output error caused by diffraction is simulated by using the experimental parameters and input data. Analysis shows that near-field diffraction causes large output errors. When the combined effects of near-field and far-field diffraction are combined, the error varies greatly due to different input image modes. The relative error is larger when a smaller simple image is input. For the complex images actually used in the experiment, the relative error is small . The output data is corrected by linear regression method and its feasibility is analyzed. The corrected data error is reduced by one order of magnitude. The influence of diffraction error on the recognition rate in the experiment can be controlled within a small range, and the recognition rate can be maintained above 97.7%.