使用YOLOv3训练一个数字串识别模型，端到端实现数字串识别

YOLOv3数字串识别

使用YOLOv3模型训练一个字符串识别模型，例如车牌、数码管、刻度值等，端到端的无字符分割的快速实现数字串的识别。
使用c++调用模型直接输出字符串数值：

#include <fstream>
#include <sstream>
#include <iostream>
#include <opencv2/dnn.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/highgui.hpp>
#include <vector>
#include <math.h>
#include<ctime>
#include<algorithm>
#include <fstream>


using namespace std;
using namespace cv;
using namespace dnn;
vector<string> classes;
vector<int> indices;
int indices_length;
//int *segnum1 = new int[indices_length];   //动态创建一个长度为roi_Final.size()即分割的字符长度的数组，用于存放识别的数字
//char *segnum2 = new char[indices_length];   //创建字符数组用于把segnum1整型转为字符型方便后面插入小数点字符

int segnum1[4];
char segnum2[4];



											//对识别的数字排序
struct node
{
	int value;
	int index;
};

bool cmp(struct node a, struct node b)
{
	if (a.value < b.value)
	{
		return true;
	}
	return false;
}

template <typename T>
T sort_indexes(vector<size_t> &idx, vector<T> &v)
{
	node* a = new node[v.size()];
	for (int i = 0; i < v.size(); i++)
	{
		a[i].value = v[i];
		a[i].index = i;
	}
	std::sort(a, a + v.size(), cmp);
	for (int i = 0; i < v.size(); i++)
	{
		idx.push_back(a[i].index);
	}
	delete[] a;

	return 0;
}


//原文链接：https ://blog.csdn.net/u013925378/article/details/83865707



vector<String> getOutputsNames(Net&net)
{
	static vector<String> names;
	if (names.empty())
	{
		//Get the indices of the output layers, i.e. the layers with unconnected outputs
		vector<int> outLayers = net.getUnconnectedOutLayers();

		//get the names of all the layers in the network
		vector<String> layersNames = net.getLayerNames();

		// Get the names of the output layers in names
		names.resize(outLayers.size());
		for (size_t i = 0; i < outLayers.size(); ++i)
			names[i] = layersNames[outLayers[i] - 1];
	}
	return names;
}


void drawPred(int classId, float conf, int left, int top, int right, int bottom, Mat& frame)
{
	//Draw a rectangle displaying the bounding box
	rectangle(frame, Point(left, top), Point(right, bottom), Scalar(0, 0, 255), 3);//矩形框大小及颜色

																				   //Get the label for the class name and its confidence
	string label = format("%.3f", conf);   //预测值保留小数点后两位
	if (!classes.empty())
	{
		CV_Assert(classId < (int)classes.size());
		label = classes[classId] + ":" + label;
	}

	//Display the label at the top of the bounding box
	int baseLine;
	Size labelSize = getTextSize(label, FONT_HERSHEY_SIMPLEX, 0, 1, &baseLine);   //0表示预测框上面的文本条大小，0表示无
	top = max(top, labelSize.height);
	rectangle(frame, Point(left, top - round(0.5*labelSize.height)), Point(left + round(0.5*labelSize.width), top + baseLine), Scalar(255, 255, 255), FILLED);
	putText(frame, label, Point(left, top), FONT_HERSHEY_SIMPLEX, 1, Scalar(255, 0, 0), 3);  //0.4表示预测字体的大小，1表示字体的粗细
																							 //putText(frame, label, Point(left, top), FONT_HERSHEY_SIMPLEX, 0.4, Scalar(255, 0, 0), 1.4);  //0.4表示预测字体的大小，1表示字体的粗细

}



void postprocess(Mat& frame, const vector<Mat>& outs, float confThreshold, float nmsThreshold)
{
	vector<int> classIds;
	vector<float> confidences;
	vector<Rect> boxes;

	for (size_t i = 0; i < outs.size(); ++i)
	{
		// Scan through all the bounding boxes output from the network and keep only the
		// ones with high confidence scores. Assign the box's class label as the class
		// with the highest score for the box.
		float* data = (float*)outs[i].data;
		for (int j = 0; j < outs[i].rows; ++j, data += outs[i].cols)
		{
			Mat scores = outs[i].row(j).colRange(5, outs[i].cols);
			Point classIdPoint;
			double confidence;
			// Get the value and location of the maximum score
			minMaxLoc(scores, 0, &confidence, 0, &classIdPoint);
			if (confidence > confThreshold)
			{
				int centerX = (int)(data[0] * frame.cols);
				int centerY = (int)(data[1] * frame.rows);
				int width = (int)(data[2] * frame.cols);
				int height = (int)(data[3] * frame.rows);
				int left = centerX - width / 2;
				int top = centerY - height / 2;

				classIds.push_back(classIdPoint.x);
				confidences.push_back((float)confidence);
				boxes.push_back(Rect(left, top, width, height));
			}
		}
	}

	// Perform non maximum suppression to eliminate redundant overlapping boxes with
	// lower confidences

	NMSBoxes(boxes, confidences, confThreshold, nmsThreshold, indices);

	vector<int>NumClassid;
	vector<int>Num_X;
	indices_length = indices.size();
	for (size_t i = 0; i < indices.size(); ++i)
	{
		int idx = indices[i];
		Rect box = boxes[idx];
		drawPred(classIds[idx], confidences[idx], box.x, box.y,
			box.x + box.width, box.y + box.height, frame);
		Num_X.push_back(box.x);
		NumClassid.push_back(classIds[idx]);
	}


	vector<size_t> idx;
	sort_indexes(idx, Num_X);
	for (int i = 0; i < NumClassid.size(); i++)
	{
		//cout << NumClassid[idx[i]] << endl;
		segnum1[i] = NumClassid[idx[i]];       //识别的数字依次存入一维数组中

	}


}

Mat numReco(string imageFile)
{


	string names_file = "D:\\PointerImg\\darknet-master-num\\data\\voc.names";
	String model_def = "D:\\PointerImg\\darknet-master-num\\cfg\\yolov3-voc.cfg";
	String weights = "D:\\PointerImg\\darknet-master-num\\backup\\yolov3-voc_last.weights";

	int in_w, in_h;
	double thresh = 0.7;
	double nms_thresh = 0.25;
	in_w = in_h = 320;

	//read names
	ifstream ifs(names_file.c_str());
	string line;
	while (getline(ifs, line)) classes.push_back(line);

	//init model
	Net net = readNetFromDarknet(model_def, weights);
	net.setPreferableBackend(DNN_BACKEND_OPENCV);
	net.setPreferableTarget(DNN_TARGET_CPU);

	//read image and forward
	VideoCapture capture(2);// VideoCapture:OENCV中新增的类，捕获视频并显示出来
	Mat  blob, frame;
	capture >> frame;

	frame = imread(imageFile);
	if (!frame.data)
	{
		cout << "输入图片不存在！" << endl;
		system("pause");
		exit(0);
	}
	blobFromImage(frame, blob, 1 / 255.0, Size(in_w, in_h), Scalar(), true, false);

	vector<Mat> mat_blob;
	imagesFromBlob(blob, mat_blob);

	//Sets the input to the network
	net.setInput(blob);

	// Runs the forward pass to get output of the output layers
	vector<Mat> outs;
	net.forward(outs, getOutputsNames(net));

	postprocess(frame, outs, thresh, nms_thresh);

	vector<double> layersTimes;
	double freq = getTickFrequency() / 1000;
	double t = net.getPerfProfile(layersTimes) / freq;
	//string label = format("Inference time for a frame : %.2f ms", t);
	string label = format("", t);
	putText(frame, label, Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 255, 0));
	//printf("Inference time for a frame : %.2f ms", t);
	imshow("num location", frame);
	//imwrite("../../data/predicts/meterLocal.jpg", frame);

	return frame;
}

double ScalesNumReco()
{
	//int型转char
	for (int i = 0; i < indices_length; i++)
	{
		segnum2[i] = char(segnum1[i] + 48);
	}
	char b = char(46);

	if (segnum2[0] == '0' && indices_length > 1)    //识别的数字字符的第一个字符为0，且字符长度大于1，则读数为x.xxx
	{
		for (int i = indices_length - 1; i != 0; i--)  //从后往前移动，防止覆盖
		{
			segnum2[i + 1] = segnum2[i];
		}
		segnum2[1] = b;
	}

	double number = atof(segnum2);   //把字符数组转化为转换成字符串s
									 //meterNumReco.push_back(number);   
	//delete segnum1, segnum2;    //释放动态数组 
								//indices.clear();
	return number;
	//string s(segnum2);//  将字符数组numStr


}

int main()
{

	clock_t startTime, endTime;
	startTime = clock();

	//String imageFile1 = "E:/picture/meter/ScaleNumReco/scale_num/*.jpg";
	String imageFile1 = "E:/picture/meter/ScaleNumReco/scale_num/022150.jpg";
	vector<cv::String> image_files;
	glob(imageFile1, image_files);
	if (image_files.size() == 0) {
		std::cout << "No image files[jpg]" << std::endl;
		return 0;
	}

	string str;
	ofstream out("E:/picture/meter/ScaleNumReco/out.txt");
	for (unsigned int frame = 0; frame < image_files.size(); ++frame)//image_file.size()代表文件中总共的图片个数
	{

		stringstream ss1, ss2;
		ss1 << frame;
		ss1 >> str;

		cout << image_files[frame] << endl;

		Mat prediction = numReco(image_files[frame]);
		double result = ScalesNumReco();
		cout << "刻度数字为：" << result << endl;
		imwrite("E:/picture/meter/ScaleNumReco/scale_num_pre/" + str + ".jpg", prediction);
		
		out << image_files[frame] << ": " << result << endl;

		//清空数组，给数组全部赋值为0
		for (int i = 0; i < 4; i++)
		{
			segnum1[i] = 0;
			segnum2[i] = 0;
		}
			

	}
	out.close();

	endTime = clock();//计时结束
	cout << "The total run time is: " << (double)(endTime - startTime) / CLOCKS_PER_SEC << "s" << endl;
	system("pause");
	return 0;

}

以仪表刻度值为例，输入图像如下：

输出结果为：