Abstract: According to the difficulties of coding strategy of Spiking Neural Network(SNN) for image segmentation, two types of Time-to-First- Spike coding methods are proposed: linear coding and non-linear coding. Linear coding uses a linear function that realizes the corresponding relationship from pixel values to the spike times of the neurons, while non-linear coding uses corresponding relationship of Sigmoid function. Experimental results of image segmentation show that, the segmentation result using the non-linear coding is better than the result using the linear coding, and the segmentation image of non-linear coding has greater Shannon entropy. The method of non-linear coding is easier to select optimal parameters, and acquires the best segmentation result of image.

Key words: Time-to-First-Spike coding, Sigmoid function, Spiking Neural Network(SNN), image segmentation, receptive field, maximum Shannon entropy

摘要： 针对脉冲神经网络图像分割中的脉冲编码问题，基于Time-to-First-Spike编码策略提出2种编码方式：线性编码和非线性编码。线性编码方法采用从图像像素值到神经元脉冲发放时间的线性函数对应关系，而非线性编码方法采用Sigmoid函数的对应关系。应用2种方法对图像进行分割，实验结果表明，非线性编码方法的分割结果优于线性编码方法，分割图像具有更大的香农熵值，并且非线性编码方法在图像分割时具有更大的取值区间，更容易对参数进行选择，取得最佳的图像分割结果。

关键词: Time-to-First-Spike编码, Sigmoid函数, 脉冲神经网络, 图像分割, 感受野, 最大香农熵

CLC Number: 

• TP391.41