BP神经网络图象边缘检测器及其内部信息

BP NEURAL NETWORK IMAGE EDGE DETECTOR AND ITS INTERNAL INFORMATION

介绍用神经网络实现图象边缘检测的实验研究结果。一幅数字图象用一个规模不大的ＢＰ网络边缘检测器处理小邻区，再用扫描此图象的方法进行边缘检测。此方法之最大优点是设计简单，网络边缘检测器的性能也令人满意．欠缺是：学习时间长，神经网络隐层中权重包含的信息难于解释清楚。

A number of experiments were carried out to investigate the application of neural networks for edge detection. An image can be detected by scanning the image,and processing each local neighbourhood with the BP network edge delector. One of the greatest advantages of using neural networks is that they are easy to design,and the performance of the neural edge detector seems to be quite acceptable. But the learning time can be very long and the internal information of the weights in the hidden layer is not easy to interpret.