基于参数可调的MSPCNN模型禁令交通标志分割方法

Segmentation Method of Forbidden Traffic Signs Based on MSPCNN Model with Adjustable Parameters

针对脉冲耦合神经网络在交通标志分割中准确度不高和参数设置复杂的问题,提出一种参数可调的改进脉冲耦合神经网络(PA-MSPCNN)。通过分析交通标志颜色特征,对图像进行红化预处理,区分出交通标志和环境背景;根据周围神经元对中心神经元的影响,改进MSPCNN模型中的加权矩阵和连接系数;通过分析动态阈值间的关系,增设辅助参数,使动态阈值的调节更加合理。实验结果表明,PA-MSPCNN对交通标志的检测准确率达85%。PA-MSPCNN在减少传统PCNN模型中的参数量同时,能准确分割图像,在光照条件变化、交通标志尺度变化和几何旋转等复杂情况下具有更好的适用性。

Aiming at the problems of low accuracy and complex parameter setting in the traffic sign segmentation of a pulse-coupled neural network,we propose an improved pulse-coupled neural network with adjustable parameters(PA-MSPCNN)in this paper.By analyzing the color characteristics of traffic signs,the PA-MSPCNN preprocesses the image with reddening and distinguishes traffic signs and the environmental background.The influence of neighboring neurons on central neurons improves the weighing matrix and the connection coefficient of the MSPCNN.We analyze the relationship between the dynamic thresholds and adjust these more reasonably by adding an auxiliary parameter.The experimental results show that the segmentation accuracy of the PA-MSPCNN on traffic sign images is 85%.The PA-MSPCNN not only reduces the number of parameters in the traditional PCNN model but also accurately segments the image,which has better applicability for complex situations such as changes in illumination conditions,scale changes,and geometric rotation of traffic signs.