由于光学固有的数值精度低，难以表示负值等弱点，用光学方法实现神经网络时存在着许多困难。针对光学的弱点，本文提出并建立了具有单极二值互连的适应截值模型，这一模型避开了光学实现时难以表示负值和互连精度差等弱点，计算机模拟及光学实现结果表明，这种单极互连神经网络模型同其他的单极模型相比具有高的存储容量及较强的寻址能力。 There are many difficulties in using optical methods to realize neural networks due to the inherent numerical accuracy of optics and the difficulty of expressing weaknesses such as negative values. In view of the optical weakness, this paper proposes and establishes an adaptive cut-off model with a unipolar binary interconnection. This model avoids the weakness of difficult to represent negative values ​​and poor interconnection accuracy in optical implementation. The results of computer simulation and optics implementation It shows that this unipolar neural network model has high storage capacity and strong addressing ability compared with other unipolar models.