chap11

说明：  本目录主要包括：文档和两个源代码。 其中一个源代码是识别程序的，另一个是一个矩阵类库的，书里面也使用过，所以一并附上。 因为已经保存了训练好的网络权值，所以第一次使用识别程序时，可以直接运行\\Release目录下的可执行文件，然后对图片目录中的测试图片进行读入、识别。 当然读者也可以自行用训练样本训练网络，不过要特别注意训练样本的选择，否则可能造成识别率很低。训练样本选择的原则是，尽可能的有代表性，在训练时间不至于太长的情况下，训练样本数目尽量多。 ( This directory includes: documentation and two source code. One of the source code is to identify the procedure, the other is a matrix class library, the book which is also used, so be attached. Because it has saved the trained network weights, so the first use of identification procedures, you can directly run \ \ Release directory of the executable files, and then tested on the images directory to read pictures, identification. Of course, readers are free to use the training samples to train the network, but pay special attention to training samples, otherwise it may cause very low recognition rate. Training sample selection principle is that as far as possible, representative, in the long training time Buzhi Yu s case, the number of training samples as much as possible. )