基于Java实现CNN，并附MNIST和语音MFCC特征性别识别。

构建CNN LayerBuilder builder = new LayerBuilder(); builder.addLayer(Layer.buildInputLayer(new Size(28, 28))); builder.addLayer(Layer.buildConvLayer(6, new Size(5, 5))); builder.addLayer(Layer.buildSampLayer(new Size(2, 2))); builder.addLayer(Layer.buildConvLayer(12, new Size(5, 5))); builder.addLayer(Layer.buildSampLayer(new Size(2, 2))); builder.addLayer(Layer.buildOutputLayer(10)); CNN cnn = new CNN(builder, 50); 运行MNIST数据集 String fileName = "data/train.format"; Dataset dataset = Dataset.load(fileName, ",", 784); cnn.train(dataset, 100); Dataset testset = Dataset.load("data/test.format", ",", -1); cnn.predict(testset, "data/test.predict"); 计算精度可以达到97.8%。

package info.hb.ccn.main;

import info.hb.cnn.core.CNN;
import info.hb.cnn.core.CNN.layerBuilder;
import info.hb.cnn.core.layer;
import info.hb.cnn.core.layer.Size;
import info.hb.cnn.data.DataSet;
import info.hb.cnn.utils.ConcurentRunner;

public class CNNMnist {

	private static final String MODEL_NAME = “mnist/model/model.cnn“;

	private static final String TRAIN_DATA = “mnist/train.format“;

	private static final String TEST_DATA = “mnist/test.format“;

	private static final String TEST_PREDICT = “mnist/test.predict“;

	public static void main（String[] args） {

		System.err.println（“训练阶段：“）;
		runTrain（）;
		System.err.println（“测试阶段：“）;
		runTest（）;
		ConcurentRunner.stop（）;

	}

	public static void runTrain（） {
		// 构建网络层次结构
		layerBuilder builder = new layerBuilder（）;
		builder.addlayer（layer.buildInputlayer（new Size（28 28）））; // 输入层输出map大小为28×28
		builder.addlayer（layer.buildConvlayer（6 new Size（5 5）））; // 卷积层输出map大小为24×2424=28+1-5
		builder.addlayer（layer.buildSamplayer（new Size（2 2）））; // 采样层输出map大小为12×1212=24/2
		builder.addlayer（layer.buildConvlayer（12 new Size（5 5）））; // 卷积层输出map大小为8×88=12+1-5
		builder.addlayer（layer.buildSamplayer（new Size（2 2）））; // 采样层输出map大小为4×44=8/2
		builder.addlayer（layer.buildOutputlayer（10））;
		CNN cnn = new CNN（builder 10）;
		// 加载训练数据
		DataSet dataset = DataSet.load（TRAIN_DATA ““ 784）;
		// 开始训练模型
		cnn.train（dataset 5）;
		// 保存训练好的模型
		cnn.saveModel（MODEL_NAME）;
		dataset.clear（）;
	}

	public static void runTest（） {
		// 加载训练好的模型
		CNN cnn = CNN.loadModel（MODEL_NAME）;
		// 加载测试数据
		DataSet testSet = DataSet.load（TEST_DATA ““ -1）;
		// 预测结果
		cnn.predict（testSet TEST_PREDICT）;
		testSet.clear（）;
	}

}

 属性            大小     日期    时间   名称
----------- ---------  ---------- -----  ----
     目录           0  2016-08-21 08:14  cnn-master\
     文件          45  2016-08-21 08:14  cnn-master\.gitignore
     文件         754  2016-08-21 08:14  cnn-master\README.md
     目录           0  2016-08-21 08:14  cnn-master\mnist\
     文件         208  2016-08-21 08:14  cnn-master\mnist\readme.md
     文件    43904000  2016-08-21 08:14  cnn-master\mnist\test.format
     文件    18841570  2016-08-21 08:14  cnn-master\mnist\train.format
     文件        3024  2016-08-21 08:14  cnn-master\pom.xml
     目录           0  2016-08-21 08:14  cnn-master\speech\
     目录           0  2016-08-21 08:14  cnn-master\speech\model\
     文件       16718  2016-08-21 08:14  cnn-master\speech\model\model.cnn
     文件       31110  2016-08-21 08:14  cnn-master\speech\test.format
     文件          86  2016-08-21 08:14  cnn-master\speech\test.label
     文件          86  2016-08-21 08:14  cnn-master\speech\test.predict
     文件      144463  2016-08-21 08:14  cnn-master\speech\train.format
     目录           0  2016-08-21 08:14  cnn-master\src\
     目录           0  2016-08-21 08:14  cnn-master\src\main\
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\hb\
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\hb\ccn\
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\hb\ccn\main\
     文件        1887  2016-08-21 08:14  cnn-master\src\main\java\info\hb\ccn\main\CNNMnist.java
     文件        1681  2016-08-21 08:14  cnn-master\src\main\java\info\hb\ccn\main\CNNSpeech.java
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\hb\cnn\
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\hb\cnn\core\
     文件       14926  2016-08-21 08:14  cnn-master\src\main\java\info\hb\cnn\core\CNN.java
     文件        5893  2016-08-21 08:14  cnn-master\src\main\java\info\hb\cnn\core\layer.java
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\hb\cnn\data\
     文件        3714  2016-08-21 08:14  cnn-master\src\main\java\info\hb\cnn\data\DataSet.java
     目录           0  2016-08-21 08:14  cnn-master\src\main\java\info\hb\cnn\utils\
............此处省略17个文件信息