摘要
基于首脉冲触发时间编码策略,提出一种应用脉冲神经网络模型进行图像分割的方法。在输入层采用首脉冲触发时间编码策略将图像的像素值转换为神经元的脉冲发放时间,将编码结果以感受野为单位送入中间层,通过阈值电位控制神经元的脉冲发放,在输出层根据分割阈值将神经元的脉冲发放时间分成2类,输出分割结果,并使用最大Shannon熵准则分析感受野大小、阈值电位和分割阈值等参数的变化对图像分割结果的影响。通过对具有噪声的复杂图像进行分割,并与最大类间方差法和基于最大熵的脉冲耦合神经网络方法进行比较,实验结果表明,该方法对噪声图像的鲁棒性较强,能获得较好的分割结果。
Using the Time-to-first-spike(TTFS) coding strategy,a novel image segmentation method of spiking neural network model is proposed.The image pixel values are encoded into the spike timings of neurons in the input layer.The encoded results are delivered into the middle layer of spiking neural network through the different receptive fields,and spikes are triggered by a threshold condition.The spike timings of the neurons in the output layer are divided into two categories by the segmentation threshold.The corresponding image segmentation experiments show that the parameter changes of receptive filed size,threshold potential and segmentation threshold have significant impacts on the image segmentation result which is evaluated by the maximum Shannon entropy.Compared with the maximum between-cluster variance method and the pulse coupled neural network method based on maximum entropy,the proposed method has stronger robustness for image noise in image segmentation.
出处
《计算机工程》
CAS
CSCD
北大核心
2015年第8期227-232,共6页
Computer Engineering
基金
国家自然科学基金资助项目(61165002)
甘肃省自然科学基金资助项目(1010RJZA019)