【算法介绍】
智慧水务中的水表读数检测与识别系统,基于先进的YOLOv8算法,为城市供水管理的智能化升级提供了坚实的技术支撑。该系统通过集成YOLOv8的深度学习框架,实现了对水表数字及关键标识的实时、精准识别,覆盖["0","1","2","3","4","5","6","7","8","9","counter","liter"]等核心类别。
YOLOv8以其卓越的检测速度与高精度特性,能够快速解析水表图像中的数字信息,并精准定位计数器区域("counter")与单位标识("liter")。通过YOLOv8算法自动提取特征信息,完成数字识别与状态判断。此外,该系统支持多场景适配与动态追踪功能,可同时监测不同型号水表的读数变化,并兼容低光照、反光等复杂环境,如果增强数据集进行训练可以进一步提升了水务管理的精细化水平。
综上所述,基于YOLOv8的水表读数检测与识别系统,为智慧水务建设注入了创新动能,有效保障了供水计量的准确性,推动了城市水资源管理的数字化转型。
【效果展示】


注意:系统采用yolov8x大模型进行训练,所以单张图片预测1.1秒左右,如果采用yolov8n训练可以达到50ms左右。由于水表方向不确定为了能够正常读取完整水表数字需要正向放置,代码采用位置从左到右提取完整数字,最终显示在软件左上角区域。水表只读取整数部分,小数部分只检测没有识别,因为这个主要是依靠目标检测进行定位识别。
【训练数据集介绍】
注意数据集中有很多增强图片,主要是旋转增强图片
数据集格式:Pascal VOC格式+YOLO格式(不包含分割路径的txt文件,仅仅包含jpg图片以及对应的VOC格式xml文件和yolo格式txt文件)
图片数量(jpg文件个数):3552
标注数量(xml文件个数):3552
标注数量(txt文件个数):3552
标注类别数:12
所在github仓库:firc-dataset
标注类别名称(注意yolo格式类别顺序不和这个对应,而以labels文件夹classes.txt为准):["0","1","2","3","4","5","6","7","8","9","counter","liter"]
每个类别标注的框数:
0 框数 = 3865
1 框数 = 1768
2 框数 = 1390
3 框数 = 1295
4 框数 = 1198
5 框数 = 1074
6 框数 = 1019
7 框数 = 827
8 框数 = 908
9 框数 = 864
counter 框数 = 3552
liter 框数 = 3551
总框数:21311
图片分辨率:640x640
使用标注工具:labelImg
标注规则:对类别进行画矩形框
重要说明:暂无
特别声明:本数据集不对训练的模型或者权重文件精度作任何保证
图片预览:


标注例子:


【测试环境】
windows10 x64系统
VS2019
netframework4.7.2
opencvsharp4.9.0
onnxruntime1.22.0
【训练信息】
|-----------------|-------|
| 参数 | 值 |
| 训练集图片数 | 3072 |
| 验证集图片数 | 480 |
| 训练map | 98.6% |
| 训练精度(Precision) | 96.5% |
| 训练召回率(Recall) | 96.6% |
【界面设计代码】
using DeploySharp.Data;
using OpenCvSharp;
using System;
using System.Collections.Generic;
using System.ComponentModel;
using System.Data;
using System.Diagnostics;
using System.Drawing;
using System.IO;
using System.Linq;
using System.Text;
using System.Threading;
using System.Threading.Tasks;
using System.Windows.Forms;
namespace FIRC
{
public partial class Form1 : Form
{
public bool videoStart = false;//视频停止标志
string weightsPath = Application.StartupPath + "\\weights";//模型目录
YoloDetector detetor = new YoloDetector();//推理引擎
public Form1()
{
InitializeComponent();
CheckForIllegalCrossThreadCalls = false;//线程更新控件不报错
}
private void LoadWeightsFromDir()
{
var di = new DirectoryInfo(weightsPath);
foreach(var fi in di.GetFiles("*.onnx"))
{
comboBox1.Items.Add(fi.Name);
}
if(comboBox1.Items.Count>0)
{
comboBox1.SelectedIndex = 0;
}
else
{
tssl_show.Text = "未找到模型,请关闭程序,放入模型到weights文件夹!";
tsb_pic.Enabled = false;
tsb_video.Enabled = false;
tsb_camera.Enabled = false;
}
}
private void Form1_Load(object sender, EventArgs e)
{
LoadWeightsFromDir();//从目录加载模型
}
public string GetResultString(DetResult[] result)
{
Dictionary<string, int> resultDict = new Dictionary<string, int>();
for (int i = 0; i < result.Length; i++)
{
if(resultDict.ContainsKey( result[i].Category) )
{
resultDict[result[i].Category]++;
}
else
{
resultDict[result[i].Category] =1;
}
}
var resultStr = "";
foreach(var item in resultDict)
{
resultStr += string.Format("{0}:{1}\r\n",item.Key,item.Value);
}
return resultStr;
}
private void tsb_pic_Click(object sender, EventArgs e)
{
OpenFileDialog ofd = new OpenFileDialog();
ofd.Filter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";
if (ofd.ShowDialog() != DialogResult.OK) return;
tssl_show.Text = "正在检测中...";
Task.Run(() => {
var sw = new Stopwatch();
sw.Start();
Mat image = Cv2.ImRead(ofd.FileName);
detetor.SetParams(Convert.ToSingle(numericUpDown1.Value), Convert.ToSingle(numericUpDown2.Value));
var results=detetor.Inference(image);
var resultImage = detetor.DrawImage(image, results);
sw.Stop();
pb_show.Image = OpenCvSharp.Extensions.BitmapConverter.ToBitmap(resultImage);
tb_res.Text = GetResultString(results);
tssl_show.Text = "检测已完成!总计耗时"+sw.Elapsed.TotalSeconds+"秒";
});
}
public void VideoProcess(string videoPath)
{
Task.Run(() => {
detetor.SetParams(Convert.ToSingle(numericUpDown1.Value), Convert.ToSingle(numericUpDown2.Value));
VideoCapture capture = new VideoCapture(videoPath);
if (!capture.IsOpened())
{
tssl_show.Text="视频打开失败!";
return;
}
Mat frame = new Mat();
var sw = new Stopwatch();
int fps = 0;
while (videoStart)
{
capture.Read(frame);
if (frame.Empty())
{
Console.WriteLine("data is empty!");
break;
}
sw.Start();
var results = detetor.Inference(frame);
var resultImg = detetor.DrawImage(frame,results);
sw.Stop();
fps = Convert.ToInt32(1 / sw.Elapsed.TotalSeconds);
sw.Reset();
Cv2.PutText(resultImg, "FPS=" + fps, new OpenCvSharp.Point(30, 30), HersheyFonts.HersheyComplex, 1.0, new Scalar(255, 0, 0), 3);
//显示结果
pb_show.Image = OpenCvSharp.Extensions.BitmapConverter.ToBitmap(resultImg);
tb_res.Text = GetResultString(results);
Thread.Sleep(5);
}
capture.Release();
pb_show.Image = null;
tssl_show.Text = "视频已停止!";
tsb_video.Text = "选择视频";
});
}
public void CameraProcess(int cameraIndex=0)
{
Task.Run(() => {
detetor.SetParams(Convert.ToSingle(numericUpDown1.Value), Convert.ToSingle(numericUpDown2.Value));
VideoCapture capture = new VideoCapture(cameraIndex);
if (!capture.IsOpened())
{
tssl_show.Text = "摄像头打开失败!";
return;
}
Mat frame = new Mat();
var sw = new Stopwatch();
int fps = 0;
while (videoStart)
{
capture.Read(frame);
if (frame.Empty())
{
Console.WriteLine("data is empty!");
break;
}
sw.Start();
var results = detetor.Inference(frame);
var resultImg = detetor.DrawImage(frame, results);
sw.Stop();
fps = Convert.ToInt32(1 / sw.Elapsed.TotalSeconds);
sw.Reset();
Cv2.PutText(resultImg, "FPS=" + fps, new OpenCvSharp.Point(30, 30), HersheyFonts.HersheyComplex, 1.0, new Scalar(255, 0, 0), 3);
//显示结果
pb_show.Image = OpenCvSharp.Extensions.BitmapConverter.ToBitmap(resultImg);
tb_res.Text = GetResultString(results);
Thread.Sleep(5);
}
capture.Release();
pb_show.Image = null;
tssl_show.Text = "摄像头已停止!";
tsb_camera.Text = "打开摄像头";
});
}
private void tsb_video_Click(object sender, EventArgs e)
{
if(tsb_video.Text=="选择视频")
{
OpenFileDialog ofd = new OpenFileDialog();
ofd.Filter = "视频文件(*.*)|*.mp4;*.avi";
if (ofd.ShowDialog() != DialogResult.OK) return;
videoStart = true;
VideoProcess(ofd.FileName);
tsb_video.Text = "停止";
tssl_show.Text = "视频正在检测中...";
}
else
{
videoStart = false;
}
}
private void tsb_camera_Click(object sender, EventArgs e)
{
if (tsb_camera.Text == "打开摄像头")
{
videoStart = true;
CameraProcess(0);
tsb_camera.Text = "停止";
tssl_show.Text = "摄像头正在检测中...";
}
else
{
videoStart = false;
}
}
private void tsb_exit_Click(object sender, EventArgs e)
{
videoStart = false;
this.Close();
}
private void trackBar1_Scroll(object sender, EventArgs e)
{
numericUpDown1.Value = Convert.ToDecimal(trackBar1.Value / 100.0f);
}
private void trackBar2_Scroll(object sender, EventArgs e)
{
numericUpDown2.Value = Convert.ToDecimal(trackBar2.Value / 100.0f);
}
private void numericUpDown1_ValueChanged(object sender, EventArgs e)
{
trackBar1.Value = (int)(Convert.ToSingle(numericUpDown1.Value) * 100);
}
private void numericUpDown2_ValueChanged(object sender, EventArgs e)
{
trackBar2.Value = (int)(Convert.ToSingle(numericUpDown2.Value) * 100);
}
private void comboBox1_SelectedIndexChanged(object sender, EventArgs e)
{
tssl_show.Text="加载模型:"+comboBox1.Text;
detetor.LoadWeights(weightsPath+"\\"+comboBox1.Text);
tssl_show.Text = "模型加载已完成!";
}
}
}
【使用步骤】
使用步骤:
(1)首先根据官方框架ultralytics安装教程安装好yolov8环境,并根据官方export命令将自己pt模型转成onnx模型,然后去github.com/futureflsl/firc-csharp-projects找到源码
(2)使用vs2019打开sln项目,选择x64 release并且修改一些必要的参数,比如输入shape等,点击运行即可查看最后效果
特别注意如果运行报错了,请参考我的博文进行重新引用我源码的DLL:[C#]opencvsharp报错System.Memory,Version=4.0.1.2,Culture=neutral,PublicKeyToken=cc7b13fcd2ddd51"版本高于所引_未能加载文件或程序集"system.memory, version=4.0.1.2, culture-CSDN博客
【提供文件】
C#源码
yolov8x.onnx模型(不提供pytorch模型)
训练的map,P,R曲线图(在weights\results.png)
测试图片(在test_img文件夹下面)
训练数据集