/// <summary>
///
/// </summary>
/// <param name="matrix_size"></param>
/// <param name="steps"></param>
/// <returns></returns>
private int[][,] ToMatrix(Size matrix_size, out List <int> steps)
{
var metrix = new List <int[, ]>();
steps = new List <int>();
for (int id_char = 0; id_char < Characters.Count; id_char++)
{
var character = Characters[id_char];
//每个图分成多个笔画,一笔一笔画到矩阵上
metrix.Add(new int[matrix_size.Width, matrix_size.Height]);
int step = 1;
foreach (var stroke in character.Strokes)
{
var side = stroke.Points.ToArray().FindSide();
//内容 与 矩阵 比例 这里是等比缩放,所以横竖轴线只取短轴的比例即可
double scale = 0;
//字符区域的大小
var charzone_size = new Size(side.Right - side.Left, side.Bottom - side.Top);
double scale_x = (double)(matrix_size.Width - 1) / (double)charzone_size.Width;
double scale_y = (double)(matrix_size.Height - 1) / (double)charzone_size.Height;
scale = scale_x < scale_y ? scale_x : scale_y;
//每个笔画分成多个像素点
for (int p = 0; p < stroke.Points.Count(); p++)
{
stroke.Points[p].X = (int)((stroke.Points[p].X - side.Left) * scale);
stroke.Points[p].Y = (int)((stroke.Points[p].Y - side.Top) * scale);
if (p == 0)
{
continue; //第一笔属于起点,跳过
}
var distance = Extends.Distance(stroke.Points[p - 1], stroke.Points[p]);
if (distance <= 0)
{
continue;
}
var ratio = (double)(stroke.Points[p].X - stroke.Points[p - 1].X) / (double)distance;
var angle = ratio.Acos();
for (int stepofstroke = 0; stepofstroke < distance; stepofstroke++)
{
int x = (int)(angle.Cos() * stepofstroke + stroke.Points[p - 1].X);
int y = (int)(angle.Sin() * stepofstroke + stroke.Points[p - 1].Y);
metrix[id_char][x, y] = step;
step += 1;
}
}
steps.Add(step);
}
}
return(metrix.ToArray());
}
static void Main(string[] args)
{
//总训练次数
const int alltimes = 10000;
Console.Title = "";
//单独读取笔画数据
var traindata = new TrainData[] {
new TrainData("Res/Strokes/S00.json", 0),
new TrainData("Res/Strokes/S01.json", 1),
new TrainData("Res/Strokes/S02.json", 2),
new TrainData("Res/Strokes/S03.json", 3),
new TrainData("Res/Strokes/S04.json", 4),
new TrainData("Res/Strokes/S05.json", 5),
new TrainData("Res/Strokes/S06.json", 6)
};
#region 合并数据到一起
var data_num = 0;
foreach (var item in traindata)
{
data_num += item.MatrixData.GetLength(0);
}
var sumdata = new TrainData();
sumdata.MatrixData = new double[data_num, traindata[0].MatrixData.GetLength(1)];
sumdata.Anwser = new double[data_num, traindata[0].Anwser.GetLength(1)];
int startpoint = 0;
for (int i = 0; i < traindata.Length; i++)
{
var item = traindata[i];
for (int k = 0; k < item.MatrixData.GetLength(0); k++)
{
for (int j = 0; j < item.MatrixData.GetLength(1); j++)
{
sumdata.MatrixData[startpoint, j] = item.MatrixData[k, j];
}
for (int j = 0; j < item.Anwser.GetLength(1); j++)
{
sumdata.Anwser[startpoint, j] = item.Anwser[k, j];
}
startpoint++;
}
}
#endregion
Console.WriteLine("加载数据完成,开始训练...");
//把"问题"和"答案"交给神经网络,让神经网络自己研究怎么把"问题"计算成"答案"
var bp = new BpNet(sumdata.MatrixData, sumdata.Anwser);
//循环进行多次训练
for (int i = 1; i < alltimes; i++)
{
bp.train(sumdata.MatrixData, sumdata.Anwser);
if (i % (alltimes / 10) == 0)
{
Console.WriteLine($"总训练次数:{alltimes} 当前训练次数:{i}");
}
}
Console.WriteLine("训练完成,开始预测..");
//加载测试数据集
var test = Extends.GetDim1Matrix("Res/test.json");
//循环把测试数据集放进去识别:
for (int i = 0; i < test.GetLength(0); i++)
{
var sim0 = bp.sim(test.GetColumn(i));
int like = 0;
var distant = double.MaxValue;
//排序.把最接近1的排到上面
for (int k = 0; k < sim0.Length; k++)
{
if (Math.Abs(sim0[k] - 1) > distant)
{
continue;
}
distant = Math.Abs(sim0[k] - 1);
like = k;
}
Console.WriteLine($"测试数据{i}识别结果:{like}");
}
var result = bp.ToString();
//把训练出来的权重数据保存到JSON
File.WriteAllText("GData.json", result);
Console.ReadKey(true);
}
