基于脉冲耦合神经网络的粘连玉米种子图像分割

为解决玉米种子内部机械裂纹检测过程中存在的种子间粘连问题,提出一种基于自适应脉冲耦合神经网络(Pulse coupled neural net)模型与熵值最大原则相结合的图像分割算法。运用直方图均衡化和布特沃斯低通滤波器进行频域增强预处理,以提高玉米种子与图像背景的对比度;运用PCNN模型,结合最大熵值原则对预处理后的粘连玉米种子图像进行分割,并引入图像像素的拉普拉斯能量(Energy of laplace)作为PCNN网络各神经元之间的连接系数,以增强图像分割效果;采用维纳滤波和数学形态学对分割后存在的噪声和断点进行处理,得到最终的分割效果。试验结果表明:PCNN与熵值最大原则相结合的图像分割算法的分割准确率为92.5%,运行时间为1.166 2s,分割准确率高于改进分水岭算法、OTSU算法和最大熵直方图分割算法,用时略长于其他分割算法。

Image segmentation of touching corn seeds based on PCNN

In order to solve the problem of image segmentation difficulty of touching seeds in inner mechanical cracks detection,a new segmentation method was proposed in this paper,which adopted image segmentation algorithms of combing adaptive pulse coupled neural net(PCNN) model and entropy maximal principle.The first,for improving contrast between touching corn seeds and image background,touching corn seeds were pretreated about enhancing the frequency,which adopted the method of histogram equalization and bout low-pass filter.The second,the image of touching corn seeds pretreated was segmented by using pulse coupled neural net(PCNN) model and combining with the principle of maximum entropy value.The energy of laplace(EOL) of image pixel was introduced as coefficient of pulse coupled neural net(PCNN) network among neurons to enhance segmentation effect of image.Finally,the final corn seeds detection result was obtained via post processing including wiener filtering and mathematical morphology breakpoint connections.The experiment result showed that:accuracy of image segmentation algorithm combining adaptive pulse coupled neural net(PCNN) model and entropy maximal principle was highest for the accuracy,at 92.5%,which was higher than the segmentation algorithm of improved watershed algorithm,OTSU algorithm and maximum entropy histogram algorithm;the running time of improved PCNN algorithm program was 1.166 2 s,which was longer than improved watershed algorithm,OTSU algorithm and maximum entropy histogram algorithm for the running time,but segmentation effect of improving PCNN algorithm program was the best.