借助OpenCv4的dnn模块
    已完成对各部分代码的注释

    1. #pragma once
    2. #include<opencv2\opencv.hpp>
    3. #include<opencv2\dnn.hpp>
    4. #include<fstream>
    5. #include<iostream>
    7. using namespace std;
    8. using namespace cv;
    9. using namespace cv::dnn;
    12. int main()
    13. {
    14.     //---------------------------------------加载类别---------------------------------------
    15.     ifstream classNamesFile("D:/OpenCV/model/coco.names");
    16.     //ifstream默认以输入方式打开文件，
    18.     vector<string> classNamesVec;
    19.     if (classNamesFile.is_open())
    20.     {
    21.         string className = "";
    22.         while (std::getline(classNamesFile, className))
    23.             classNamesVec.push_back(className);
    24.         //getline()中读入结束的回车后，结束符不放入缓存区，会将读入的\n直接去除
    25.     }
    26.     //输出网络内容
    27.     for (int i = 0; i < classNamesVec.size(); i++) {
    28.         cout << i + 1 << "\t" << classNamesVec[i].c_str() << endl;
    29.     }
    31.     //---------------------------------------模型设置---------------------------------------
    32.     //训练好的模型路径
    33.     String cfg = "D:/OpenCV/model/yolov4.cfg";
    34.     String weight = "D:/OpenCV/model/yolov4.weights";
    36.     //模型读入
    37.     dnn::Net net = readNetFromDarknet(cfg, weight);
    39.     //预处理读取的图像，并将图像读入网络
    40.     Mat frame = imread("D:/OpenCV/data/faces.jpg");
    41.     imshow("【输入图像】", frame);
    42.     //规范化
    43.     Mat inputBlob = blobFromImage(frame, 1.0 / 255, Size(608, 608), Scalar());
    44.     net.setInput(inputBlob);
    46.     //获取未连接输出层，用于网络的向前计算
    47.     std::vector<String> outNames = net.getUnconnectedOutLayersNames();
    48.     for (int i = 0; i < outNames.size(); i++) {
    49.         cout << "output layer name : " << outNames[i].c_str() << endl;
    50.     }
    51.     //完成神经网络向前计算
    52.     std::vector<Mat> outs;
    53.     net.forward(outs, outNames);
    55.     //---------------------------------------目标检测---------------------------------------
    56.     //需要的变量
    57.     float* data;
    58.     Mat scores;
    60.     vector<Rect> boxes;
    61.     vector<int> classIds;
    62.     vector<float> confidences;
    63.     int centerX, centerY, width, height, left, top;
    65.     float confidenceThreshold = 0.2;
    66.     double confidence;
    68.     Point classIdPoint;
    71.     //找出所有的目标及其位置
    72.     for (size_t i = 0; i < outs.size(); ++i) {
    73.         data = (float*)outs[i].data;
    74.         for (int j = 0; j < outs[i].rows; ++j, data += outs[i].cols) {
    75.             scores = outs[i].row(j).colRange(5, outs[i].cols);
    76.             minMaxLoc(scores, 0, &confidence, 0, &classIdPoint);
    77.             if (confidence > confidenceThreshold) {
    78.                 centerX = (int)(data[0] * frame.cols);
    79.                 centerY = (int)(data[1] * frame.rows);
    80.                 width = (int)(data[2] * frame.cols);
    81.                 height = (int)(data[3] * frame.rows);
    82.                 left = centerX - width / 2;
    83.                 top = centerY - height / 2;
    85.                 classIds.push_back(classIdPoint.x);
    86.                 confidences.push_back((float)confidence);
    87.                 boxes.push_back(Rect(left, top, width, height));
    88.             }
    89.         }
    90.     }
    91.     //执行非最大抑制，
    92.     //去掉重叠率大于nmsThreshold的候选框
    93.     vector<int> indices;
    94.     NMSBoxes(boxes, confidences, 0.3, 0.2, indices);
    97.     //---------------------------------------效果展示---------------------------------------
    99.     Scalar rectColor, textColor; //box 和 text 的颜色
    100.     Rect box, textBox;
    101.     int idx;
    102.     String className;
    103.     Size labelSize;
    105.     for (size_t i = 0; i < indices.size(); ++i) {
    106.         idx = indices[i];
    107.         className = classNamesVec[classIds[idx]];
    109.         labelSize = getTextSize(className, FONT_HERSHEY_SIMPLEX, 0.5, 1, 0);
    110.         box = boxes[idx];
    111.         textBox = Rect(Point(box.x - 1, box.y),
    112.             Point(box.x + labelSize.width, box.y - labelSize.height));
    113.         rectColor = Scalar(idx * 11 % 256, idx * 22 % 256, idx * 33 % 256);
    114.         textColor = Scalar(255 - idx * 11 % 256, 255 - idx * 22 % 256, 255 - idx * 33 % 256);
    116.         rectangle(frame, box, rectColor, 2, 8, 0);
    117.         rectangle(frame, textBox, rectColor, -1, 8, 0);
    118.         putText(frame, className.c_str(), Point(box.x, box.y - 2), FONT_HERSHEY_SIMPLEX, 0.5, textColor, 1, 8);
    119.     }
    121.     imshow("【OpenCV - YOLO】", frame);
    122.     waitKey(0);
    124.     return 0;
    126. }

    海参识别

    1. using namespace std;
    2. using namespace cv;
    3. using namespace cv::dnn;
    6. int main()
    7. {
    8.     //---------------------------------------加载类别---------------------------------------
    9.     ifstream classNamesFile("D:/OpenCV/model/yoloA/obj.names");
    10.     //ifstream默认以输入方式打开文件，
    12.     vector<string> classNamesVec;
    13.     if (classNamesFile.is_open())
    14.     {
    15.         string className = "";
    16.         while (std::getline(classNamesFile, className))
    17.             classNamesVec.push_back(className);
    18.         //getline()中读入结束的回车后，结束符不放入缓存区，会将读入的\n直接去除
    19.     }
    20.     //输出网络内容
    21.     for (int i = 0; i < classNamesVec.size(); i++) {
    22.         cout << i + 1 << "\t" << classNamesVec[i].c_str() << endl;
    23.     }
    25.     //---------------------------------------模型设置---------------------------------------
    26.     //训练好的模型路径
    27.     String cfg = "D:/OpenCV/model/yoloA/yolov4-tiny-obj.cfg";
    28.     String weight = "D:/OpenCV/model/yoloA/yolov4-tiny-obj_7000.weights";
    30.     //模型读入
    31.     dnn::Net net = readNetFromDarknet(cfg, weight);
    33.     //预处理读取的图像，并将图像读入网络
    34.     Mat frame = imread("D:/OpenCV/image/haishen.jpg");
    35.     imshow("【输入图像】", frame);
    36.     //规范化
    37.     Mat inputBlob = blobFromImage(frame, 1.0 / 255, Size(608, 608), Scalar());
    38.     net.setInput(inputBlob);
    40.     //获取未连接输出层，用于网络的向前计算
    41.     std::vector<String> outNames = net.getUnconnectedOutLayersNames();
    42.     for (int i = 0; i < outNames.size(); i++) {
    43.         cout << "output layer name : " << outNames[i].c_str() << endl;
    44.     }
    45.     //完成神经网络向前计算
    46.     std::vector<Mat> outs;
    47.     net.forward(outs, outNames);
    49.     //---------------------------------------目标检测---------------------------------------
    50.     //需要的变量
    51.     float* data;
    52.     Mat scores;
    54.     vector<Rect> boxes;
    55.     vector<int> classIds;
    56.     vector<float> confidences;
    57.     int centerX, centerY, width, height, left, top;
    59.     float confidenceThreshold = 0.2;
    60.     double confidence;
    62.     Point classIdPoint;
    65.     //找出所有的目标及其位置
    66.     for (size_t i = 0; i < outs.size(); ++i) {
    67.         data = (float*)outs[i].data;
    68.         for (int j = 0; j < outs[i].rows; ++j, data += outs[i].cols) {
    69.             scores = outs[i].row(j).colRange(5, outs[i].cols);
    70.             minMaxLoc(scores, 0, &confidence, 0, &classIdPoint);
    71.             if (confidence > confidenceThreshold) {
    72.                 centerX = (int)(data[0] * frame.cols);
    73.                 centerY = (int)(data[1] * frame.rows);
    74.                 width = (int)(data[2] * frame.cols);
    75.                 height = (int)(data[3] * frame.rows);
    76.                 left = centerX - width / 2;
    77.                 top = centerY - height / 2;
    79.                 classIds.push_back(classIdPoint.x);
    80.                 confidences.push_back((float)confidence);
    81.                 boxes.push_back(Rect(left, top, width, height));
    82.             }
    83.         }
    84.     }
    85.     //执行非最大抑制，
    86.     //去掉重叠率大于nmsThreshold的候选框
    87.     vector<int> indices;
    88.     NMSBoxes(boxes, confidences, 0.3, 0.2, indices);
    91.     //---------------------------------------效果展示---------------------------------------
    93.     Scalar rectColor, textColor; //box 和 text 的颜色
    94.     Rect box, textBox;
    95.     int idx;
    96.     String className;
    97.     Size labelSize;
    99.     for (size_t i = 0; i < indices.size(); ++i) {
    100.         idx = indices[i];
    101.         className = classNamesVec[classIds[idx]];
    103.         labelSize = getTextSize(className, FONT_HERSHEY_SIMPLEX, 0.5, 1, 0);
    104.         box = boxes[idx];
    105.         textBox = Rect(Point(box.x - 1, box.y),
    106.             Point(box.x + labelSize.width, box.y - labelSize.height));
    107.         rectColor = Scalar(idx * 11 % 256, idx * 22 % 256, idx * 33 % 256);
    108.         textColor = Scalar(255 - idx * 11 % 256, 255 - idx * 22 % 256, 255 - idx * 33 % 256);
    110.         rectangle(frame, box, rectColor, 2, 8, 0);
    111.         rectangle(frame, textBox, rectColor, -1, 8, 0);
    112.         putText(frame, className.c_str(), Point(box.x, box.y - 2), FONT_HERSHEY_SIMPLEX, 0.5, textColor, 1, 8);
    113.     }
    115.     imshow("【OpenCV - YOLO】", frame);
    116.     waitKey(0);
    118.     return 0;
    120. }