# Java深度学习

**Repository Path**: Mint-Space/DeepLearning

## Basic Information

- **Project Name**: Java深度学习
- **Description**: Deeplearning for Java.Build multi-layer perceptron neural network with JAVA.Identify the MNIST handwritten digital datasets.The accuracy rate is around 80%
- **Primary Language**: Java
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 4
- **Forks**: 1
- **Created**: 2024-04-29
- **Last Updated**: 2025-07-23

## Categories & Tags

**Categories**: Uncategorized

**Tags**: Java, DeepLearning, 神经网络, 深度学习

## README

## **Multi-layer Perception NeuralNetwork**
# 多层感知器神经网络
使用Java类构建 Multi-layer Perception NeuralNetwork

```java
public static void main(String[] args) {
        
        MultilayeredPerceptronNeuralNetworks multilayeredPerceptronNeuralNetworks = new MultilayeredPerceptronNeuralNetworks();
        multilayeredPerceptronNeuralNetworks.run();
    
        X = MnistRead.getImagesBinarization(MnistRead.TRAIN_IMAGES_FILE);
        XT = MnistRead.getImagesBinarization(MnistRead.TEST_IMAGES_FILE);

        Y = MnistRead.getLabels(MnistRead.TRAIN_LABELS_FILE);
        YT = MnistRead.getLabels(MnistRead.TEST_LABELS_FILE);
        

        //下面是CNN训练
        ConvolutionLayer cl1 = new ConvolutionLayer();
        cl1.setConvolutionNeuralNumber(6);
        cl1.setConvolutionLayerParameter(5, 1, 6, ActivationType.RELU);
        cl1.setPoolingLayerParameter(2, 2, 0, PoolingType.MaxPooling);
        ConvolutionLayer cl2 = new ConvolutionLayer();
        cl2.setConvolutionNeuralNumber(16);
        cl2.setConvolutionLayerParameter(5, 1, 2, ActivationType.RELU);
        cl2.setPoolingLayerParameter(2, 2, 0, PoolingType.MaxPooling);
        ConvolutionLayer cl3 = new ConvolutionLayer();
        cl3.setConvolutionNeuralNumber(18);
        cl3.setConvolutionLayerParameter(3, 1, 1, ActivationType.RELU);
        cl3.setPoolingLayerParameter(2, 2, 0, PoolingType.MaxPooling);

        Layer layer1 = new Layer(120, ActivationType.Sigmoid);
        Layer layer2 = new Layer(84, ActivationType.Sigmoid);
        Layer layer3 = new Layer(10, ActivationType.Sigmoid);

        ConvolutionLayerConfiguration clc = new ConvolutionLayerConfiguration(X, Y);
        clc.convolutionLayerConfiguration(cl1);
        clc.convolutionLayerConfiguration(cl2);
        clc.convolutionLayerConfiguration(cl3);
        clc.fullyConnectedLayerConfiguration(layer1);
        clc.fullyConnectedLayerConfiguration(layer2);
        clc.fullyConnectedLayerConfiguration(layer3);
        clc.learningRateConfiguration(0.05);
        clc.lossConfiguration(LossType.MeanSquareErrorLossFunction);
        clc.saveParameter("src/main/resources/DataSet/CNN.mode");
        clc.train();

        //下面是CNN模型读取运行 （可以学到90%）
        ConvolutionLayerConfiguration clc1 = new ConvolutionLayerConfiguration(XT, YT);
        clc1.readParameter("src/main/resources/DataSet/CNN.mode");
        clc1.run();
        
        }
```

### 个人网站 

[墨染随想](https://morn.ink) https://morn.ink

创建感知器网络层和感知器网络， 设置数据集和学习率以及误差函数， train就开始训练了。

激活函数有:

    Sigmoid
    RELU
    Softmax
    Tanh

损失函数有:

    MeanSquareErrorLossFunction
    L2LossFunction
    L1LossFunction
    SmoothL1LossFunction
    HuBerLossFunction  未实现
    KullbackLeiblerDivergenceFunction
    CrossEntropyLossFunction
    SoftmaxLossFunction
    FocalLossFunction  未实现


训练准确率为84.3% ^_^  
CNN 卷积神经网络准确率提升90%;


项目地址：https://gitee.com/Mint-Space/DeepLearning

项目打包: mvn -X clean package

路径 /src/main/resources/META-INF/native-image/ink.mint/deeplearning 下文件由以下命令生成

步骤 4：重新生成配置文件（推荐）
通过 GraalVM Agent 自动生成配置：

```BASH
- 1. 运行应用并生成配置  
    java -agentlib:native-image-agent=config-output-dir=./config -jar ./deeplearning.jar

- 2. 复制生成的配置到项目
  cp -r config/* src/main/resources/META-INF/native-image/



    mvn clean package -X
```

[多层感知器神经网络](https://gitee.com/Mint-Space/DeepLearning)

![微信支付宝](src/main/resources/DataSet/gitee.jpg)

![人工智能](https://pic1.zhimg.com/v2-f672a345fd1382a1c2954741ccd5a75d_r.jpg)

喜欢这个项目可以支持一下

模型正在优化中。。。。