提速阉割版卷积

src/main/java/org/wlld/App.java

@@ -18,8 +18,8 @@ import java.util.Map;
 public class App {
     public static void main(String[] args) throws Exception {
         //createNerveTest();
-        //testPic();
+        testPic();
-        test();
+        //test();
     }
 
     public static void testPic() throws Exception {
@@ -34,7 +34,7 @@ public class App {
         Map<Integer, Double> wrongTagging = new HashMap<>();//分类标注
         rightTagging.put(1, 1.0);
         wrongTagging.put(1, 0.0);
-        for (int i = 1; i < 1000; i++) {
+        for (int i = 1; i < 999; i++) {
             System.out.println("开始学习1==" + i);
             //读取本地URL地址图片,并转化成矩阵
             Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/c/c" + i + ".png");
@@ -51,9 +51,9 @@ public class App {
         ModelParameter modelParameter = templeConfig.getModel();
         String model = JSON.toJSONString(modelParameter);
         System.out.println("一次提取：" + model);
+        ModelParameter modelParameter1 = JSONObject.parseObject(model, ModelParameter.class);
 
         TempleConfig templeConfig1 = getTemple(false);
-        ModelParameter modelParameter1 = JSONObject.parseObject(model, ModelParameter.class);
         templeConfig1.insertModel(modelParameter1);
         ModelParameter modelParameter2 = templeConfig1.getModel();
         String model2 = JSON.toJSONString(modelParameter2);
@@ -63,13 +63,8 @@ public class App {
         Matrix right = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/test/a101.png");
         Matrix wrong = picture.getImageMatrixByLocal("/Users/lidapeng/Desktop/myDocment/b/b1000.png");
         //进行图像检测
-        operation1.look(right, 2);
         operation1.look(wrong, 3);
-    }
+        operation1.look(right, 2);
-
-    public static void test() throws Exception {
-        TempleConfig templeConfig = new TempleConfig();
-        templeConfig.initConvolutionalVision(true, 3024, 4032);
     }
 
     public static TempleConfig getTemple(boolean isFirst) throws Exception {

src/main/java/org/wlld/imageRecognition/Convolution.java

@@ -3,6 +3,7 @@ package org.wlld.imageRecognition;
 import org.wlld.MatrixTools.Matrix;
 import org.wlld.MatrixTools.MatrixOperation;
 import org.wlld.config.Kernel;
+import org.wlld.tools.ArithUtil;
 
 /**
  * @author lidapeng
@@ -33,7 +34,6 @@ public class Convolution {
         }
         return late(myMatrix);
     }
-
     public Matrix late(Matrix matrix) throws Exception {//迟化处理
         int xn = matrix.getX();
         int yn = matrix.getY();

src/main/java/org/wlld/imageRecognition/Operation.java

@@ -2,6 +2,7 @@ package org.wlld.imageRecognition;
 
 
 import org.wlld.MatrixTools.Matrix;
+import org.wlld.config.Kernel;
 import org.wlld.nerveEntity.SensoryNerve;
 import org.wlld.tools.ArithUtil;
 
@@ -35,14 +36,6 @@ public class Operation {//进行计算
         intoNerve(1, list, templeConfig.getSensoryNerves(), true, tagging);
     }
 
-    //卷积网络学习
-    public void learning(Matrix matrix, double tag) throws Exception {//前者是图片，后者是标注
-        //intoNerve(1, list, templeConfig.getSensoryNerves(), true, tagging);
-        intoNerveMatrix(1, matrix,
-                templeConfig.getConvolutionalNerveManager().getSensoryNerves()
-                , true, tag);
-    }
-
     //图像视觉
     public void look(Matrix matrix, long eventId) throws Exception {
         List<Double> list = convolution(matrix);
@@ -73,12 +66,4 @@ public class Operation {//进行计算
             sensoryNerveList.get(i).postMessage(eventId, featurList.get(i), isStudy, map);
         }
     }
-
-    private void intoNerveMatrix(long eventId, Matrix matrix,
-                                 List<SensoryNerve> sensoryNerveList
-            , boolean isStudy, double E) throws Exception {
-        for (int i = 0; i < sensoryNerveList.size(); i++) {
-            sensoryNerveList.get(i).postMatrixMessage(eventId, matrix, isStudy, E);
-        }
-    }
 }

src/main/java/org/wlld/imageRecognition/TempleConfig.java

@@ -1,6 +1,5 @@
 package org.wlld.imageRecognition;
 
-import org.wlld.function.ReLu;
 import org.wlld.function.Sigmod;
 import org.wlld.i.OutBack;
 import org.wlld.nerveCenter.NerveManager;
@@ -12,7 +11,6 @@ import java.util.List;
 
 public class TempleConfig {
     private NerveManager nerveManager;//神经网络管理器
-    private NerveManager convolutionalNerveManager;//卷积网络管理器
     public double cutThreshold = 10;//切割阈值默认值
     public int row = 5;//行的最小比例
     public int column = 3;//列的最小比例
@@ -31,28 +29,6 @@ public class TempleConfig {
         }
     }
 
-    public NerveManager getConvolutionalNerveManager() {
-        return convolutionalNerveManager;
-    }
-
-    public void initConvolutionalVision(boolean initPower, int weight, int height) throws Exception {//初始化卷积视觉
-        //weight 是宽度 是 j 是 cloumn , height 是 高度， row 是 i
-        //我先要确定卷积层有几层
-        int deeps = 0;
-        boolean isNext;
-        do {
-            height = height / 3;
-            weight = weight / 3;
-            isNext = height > 3 && weight > 3;
-            if (isNext) {
-                deeps++;
-            }
-        } while (isNext);
-        convolutionalNerveManager = new NerveManager(1, 1
-                , 1, deeps - 1, new ReLu());
-        convolutionalNerveManager.init(initPower, true);
-    }
-
     public ModelParameter getModel() {//获取模型参数
         return nerveManager.getModelParameter();
     }

src/main/java/org/wlld/nerveCenter/NerveManager.java

@@ -168,7 +168,7 @@ public class NerveManager {
         List<Nerve> lastNeveList = depthNerves.get(depthNerves.size() - 1);
         //初始化输出神经元
         for (int i = 1; i < outNerveNub + 1; i++) {
-            OutNerve outNerve = new OutNerve(i, hiddenNerverNub, 0, studyPoint, initPower, activeFunction, isMatrix);
+            OutNerve outNerve = new OutNerve(i, hiddenNerverNub, 0, studyPoint, initPower, activeFunction);
             //输出层神经元连接最后一层隐层神经元
             outNerve.connectFathor(lastNeveList);
             outNevers.add(outNerve);
@@ -204,8 +204,7 @@ public class NerveManager {
                 } else {
                     downNub = hiddenNerverNub;
                 }
-                HiddenNerve hiddenNerve = new HiddenNerve(j, i + 1, upNub, downNub, studyPoint, initPower, activeFunction
+                HiddenNerve hiddenNerve = new HiddenNerve(j, i + 1, upNub, downNub, studyPoint, initPower, activeFunction);
-                        , isMatrix);
                 hiddenNerveList.add(hiddenNerve);
             }
             depthNerves.add(hiddenNerveList);

src/main/java/org/wlld/nerveEntity/HiddenNerve.java

@@ -16,8 +16,8 @@ public class HiddenNerve extends Nerve {
     //static final Logger logger = LogManager.getLogger(HiddenNerve.class);
 
     public HiddenNerve(int id, int depth, int upNub, int downNub, double studyPoint,
-                       boolean init, ActiveFunction activeFunction, boolean isMatrix) {//隐层神经元
+                       boolean init, ActiveFunction activeFunction) {//隐层神经元
-        super(id, upNub, "HiddenNerve", downNub, studyPoint, init, activeFunction, isMatrix);
+        super(id, upNub, "HiddenNerve", downNub, studyPoint, init, activeFunction);
         this.depth = depth;
     }
 
@@ -40,18 +40,4 @@ public class HiddenNerve extends Nerve {
         }
         // sendMessage();
     }
-
-    @Override
-    protected void inputMatrix(long eventId, Matrix parameter, boolean isStudy
-            , double E) throws Exception {
-        //先将输入矩阵进行卷积运算
-        if (isStudy) {//初始化求和参数
-            initFeatures(eventId);
-        }
-        Matrix matrix = convolution(parameter, eventId, isStudy);//这个神经元卷积结束
-        if (isStudy) {//求和
-            outNub = calculation(eventId);
-        }
-        sendMatrixMessage(eventId, matrix, isStudy, E);//将矩阵发到下一层神经元
-    }
 }

src/main/java/org/wlld/nerveEntity/Nerve.java

@@ -1,9 +1,7 @@
 package org.wlld.nerveEntity;
 
-import org.wlld.MatrixTools.Matrix;
+
-import org.wlld.MatrixTools.MatrixOperation;
 import org.wlld.i.ActiveFunction;
-import org.wlld.test.Ma;
 import org.wlld.tools.ArithUtil;
 
 import java.util.*;
@@ -50,21 +48,14 @@ public abstract class Nerve {
     }
 
     protected Nerve(int id, int upNub, String name, int downNub,
-                    double studyPoint, boolean init, ActiveFunction activeFunction
+                    double studyPoint, boolean init, ActiveFunction activeFunction) {//该神经元在同层神经元中的编号
-            , boolean isMatrix) {//该神经元在同层神经元中的编号
         this.id = id;
         this.upNub = upNub;
         this.name = name;
         this.downNub = downNub;
         this.studyPoint = studyPoint;
         this.activeFunction = activeFunction;
-        if (init) {
+        initPower(init);//生成随机权重
-            if (isMatrix) {
-                initKernel();
-            } else {
-                initPower();//生成随机权重
-            }
-        }
     }
 
     public void sendMessage(long enevtId, double parameter, boolean isStudy, Map<Integer, Double> E) throws Exception {
@@ -77,18 +68,6 @@ public abstract class Nerve {
         }
     }
 
-    //正向传播矩阵
-    public void sendMatrixMessage(long enevtId, Matrix parameter, boolean isStudy,
-                                  double E) throws Exception {
-        if (son.size() > 0) {
-            for (Nerve nerve : son) {
-                nerve.inputMatrix(enevtId, parameter, isStudy, E);
-            }
-        } else {
-            throw new Exception("this layer is lastIndex");
-        }
-    }
-
     private void backSendMessage(long eventId) throws Exception {//反向传播
         if (fathor.size() > 0) {
             for (int i = 0; i < fathor.size(); i++) {
@@ -102,12 +81,6 @@ public abstract class Nerve {
 
     }
 
-    //接收矩阵参数
-    protected void inputMatrix(long eventId, Matrix parameter, boolean isStudy
-            , double E) throws Exception {//输入
-
-    }
-
     private void backGetMessage(double parameter, long eventId) throws Exception {//反向传播
         backNub++;
         sigmaW = ArithUtil.add(sigmaW, parameter);
@@ -135,6 +108,7 @@ public abstract class Nerve {
             double wp = ArithUtil.mul(bn, h);//编号为KEY的上层隐层神经元权重的变化值
             w = ArithUtil.add(w, wp);//修正后的编号为KEY的上层隐层神经元权重
             double dm = ArithUtil.mul(w, gradient);//返回给相对应的神经元
+            // System.out.println("allG==" + allG + ",dm==" + dm);
             wg.put(key, dm);//保存上一层权重与梯度的积
             dendrites.put(key, w);//保存修正结果
         }
@@ -157,14 +131,6 @@ public abstract class Nerve {
         return allReady;
     }
 
-    protected void initFeatures(long eventId) {//初始化九个参数和
-        List<Double> list = new ArrayList<>();
-        for (int i = 0; i < 9; i++) {
-            list.add(0.0);
-        }
-        features.put(eventId, list);
-    }
-
     protected void destoryParameter(long eventId) {//销毁参数
         features.remove(eventId);
     }
@@ -183,73 +149,23 @@ public abstract class Nerve {
         return ArithUtil.sub(sigma, threshold);
     }
 
-    //进行卷积运算
+    private void initPower(boolean init) {//初始化权重及阈值
-    protected Matrix convolution(Matrix matrix, long eventId, boolean isStudy) throws Exception {
-        Matrix kernel = getKernel();
-        int xn = matrix.getX();
-        int yn = matrix.getY();
-        int x = xn / 3;//求导后矩阵的行数
-        int y = yn / 3;//求导后矩阵的列数
-        Matrix myMatrix = new Matrix(x, y);//最终合成矩阵
-        for (int i = 0; i < xn - 3; i += 3) {//遍历行
-            for (int j = 0; j < yn - 3; j += 3) {//遍历每行的列
-                //进行卷积运算
-                double dm = MatrixOperation.convolution(matrix, kernel, i, j);
-                if (isStudy) {//如果是学习的话，拿到分块矩阵,对对应权重的值进行求和
-                    Matrix matrix1 = matrix.getSonOfMatrix(i, j, 3, 3);
-                    sigma(matrix1, eventId);
-                }
-                if (dm > 0) {//存在边缘
-                    myMatrix.setNub(i, j, dm);
-                }
-            }
-        }
-        return myMatrix;
-    }
-
-    //对每一项进行求和，只有在学习的时候执行它
-    private void sigma(Matrix matrix, long eventId) throws Exception {
-        int n = 0;
-        List<Double> list = features.get(eventId);
-        for (int i = 0; i < 3; i++) {
-            for (int j = 0; j < 3; j++) {
-                list.set(n, ArithUtil.add(list.get(n), matrix.getNumber(i, j)));
-                n++;
-            }
-        }
-    }
-
-    private Matrix getKernel() throws Exception {
-        Matrix kernel = new Matrix(3, 3);
-        for (int i = 1; i < 10; i++) {
-            double w = dendrites.get(i);
-            int t = i - 1;
-            //将权重填入卷积核当中
-            int x = t / 3;
-            int y = t % 3;
-            kernel.setNub(x, y, w);
-        }
-        return kernel;
-    }
-
-    private void initPower() {//初始化权重及阈值
         if (upNub > 0) {
             Random random = new Random();
             for (int i = 1; i < upNub + 1; i++) {
-                dendrites.put(i, random.nextDouble());//random.nextDouble()
+                double nub = 0;
+                if (init) {
+                    nub = random.nextDouble();
                 }
-            //生成随机阈值
+                dendrites.put(i, nub);//random.nextDouble()
-            threshold = random.nextDouble();
             }
+            //生成随机阈值
+            double nub = 0;
+            if (init) {
+                nub = random.nextDouble();
             }
-
+            threshold = nub;
-    public void initKernel() {//对卷积核进行初始化
-        Random random = new Random();
-        for (int i = 1; i < 10; i++) {
-            dendrites.put(i, random.nextDouble());
         }
-        //生成随机阈值
-        threshold = random.nextDouble();
     }
 
     public int getId() {

src/main/java/org/wlld/nerveEntity/OutNerve.java

@@ -19,8 +19,8 @@ public class OutNerve extends Nerve {
     private double allE;//训练累计EK
 
     public OutNerve(int id, int upNub, int downNub, double studyPoint, boolean init,
-                    ActiveFunction activeFunction, boolean isMatrix) {
+                    ActiveFunction activeFunction) {
-        super(id, upNub, "OutNerve", downNub, studyPoint, init, activeFunction, isMatrix);
+        super(id, upNub, "OutNerve", downNub, studyPoint, init, activeFunction);
     }
 
     public void setOutBack(OutBack outBack) {
@@ -54,27 +54,9 @@ public class OutNerve extends Nerve {
         }
     }
 
-    @Override
-    protected void inputMatrix(long eventId, Matrix parameter, boolean isStudy, double E) throws Exception {
-        if (isStudy) {//初始化求和参数
-            initFeatures(eventId);
-        }
-        Matrix matrix = convolution(parameter, eventId, isStudy);//这个神经元卷积结束
-        if (isStudy) {//求和
-            outNub = calculation(eventId);
-            //进行反向传播
-
-        } else {//进入BP网络输出
-
-        }
-    }
-
     private double outGradient() {//生成输出层神经元梯度变化
         //上层神经元输入值 * 当前神经元梯度*学习率 =该上层输入的神经元权重变化
         //当前梯度神经元梯度变化 *学习旅 * -1 = 当前神经元阈值变化
-        //ArithUtil.sub(E, outNub) 求这个的累计平均值
-        //allE = ArithUtil.add(ArithUtil.sub(E, outNub), allE);
-        //double avg = ArithUtil.div(allE, trainNub);
         return ArithUtil.mul(activeFunction.functionG(outNub), ArithUtil.sub(E, outNub));
     }
 }

src/main/java/org/wlld/nerveEntity/SensoryNerve.java

@@ -13,7 +13,7 @@ import java.util.Map;
 public class SensoryNerve extends Nerve {
 
     public SensoryNerve(int id, int upNub) {
-        super(id, upNub, "SensoryNerve", 0, 0.1, false, null, false);
+        super(id, upNub, "SensoryNerve", 0, 0.1, false, null);
     }
 
     public void postMessage(long eventId, double parameter, boolean isStudy, Map<Integer, Double> E) throws Exception {//感知神经元输出
@@ -21,10 +21,6 @@ public class SensoryNerve extends Nerve {
         sendMessage(eventId, parameter, isStudy, E);
     }
 
-    public void postMatrixMessage(long enevtId, Matrix parameter, boolean isStudy, double E) throws Exception {
-        sendMatrixMessage(enevtId, parameter, isStudy, E);
-    }
-
     @Override
     public void connect(List<Nerve> nerveList) {//连接第一层隐层神经元
         super.connect(nerveList);

src/main/java/org/wlld/test/Test.java

@@ -7,7 +7,10 @@ package org.wlld.test;
  */
 public class Test {
     public static void main(String[] args) {
-        int a = 1 % 3;
+
-        System.out.println(a);
+    }
+
+    public static void test1() {
+
     }
 }