一.部署流程

• 先參考上一篇博客,學(xué)會Lenet在caffe上訓(xùn)練
• caffe+qt+lenet實現(xiàn)手寫體識別源碼下載地址.
• 把訓(xùn)練好的權(quán)重文件和深度神經(jīng)網(wǎng)絡(luò)部署到Qt上,實現(xiàn)手寫體的識別.

• 環(huán)境要求(*如果未安裝請參考我的其他博客,有具體介紹)

  • Ubuntu
  • openCV
  • QtCreator
  • caffe

• 部署流程圖

開始點擊選擇圖片按鈕,讀取一張圖片圖片灰度化處理,設(shè)置網(wǎng)絡(luò)的輸入大小.圖片輸入Lenet深度卷積神經(jīng)網(wǎng)絡(luò)獲得softMax層的值獲得softMax的最大值下標結(jié)束

• 簡單界面設(shè)計?
  

代碼詳細介紹

• 代碼文件

  • Lenet-5.pro
  • lenet.h
  • mainwindow.h
  • lenet.cpp
  • main.cpp
  • mainwindow.cpp
  • mainwindow.ui

• Lenet-5.pro,配置文件.添加opencv的庫目錄和caffe庫目錄及相關(guān)依賴項

#------------------------------------------------- # # Project created by QtCreator 2018-01-11T23:43:50 # #-------------------------------------------------QT += core guigreaterThan(QT_MAJOR_VERSION, 4): QT += widgetsTARGET = Lenet-5 TEMPLATE = appSOURCES += main.cpp\mainwindow.cpp \lenet.cppHEADERS += mainwindow.h \lenet.hFORMS += mainwindow.ui#下面代碼為具體配置,路徑根據(jù)自己的修改 INCLUDEPATH += /usr/local/include \ /usr/local/include/opencv \ /usr/local/include/opencv2 LIBS += /usr/local/lib/libopencv_calib3d.so \ /usr/local/lib/libopencv_core.so \ /usr/local/lib/libopencv_features2d.so \ /usr/local/lib/libopencv_flann.so \ /usr/local/lib/libopencv_highgui.so \ /usr/local/lib/libopencv_imgcodecs.so \ /usr/local/lib/libopencv_imgproc.so \ /usr/local/lib/libopencv_ml.so \ /usr/local/lib/libopencv_objdetect.so \ /usr/local/lib/libopencv_photo.so \ /usr/local/lib/libopencv_shape.so \ /usr/local/lib/libopencv_stitching.so \ /usr/local/lib/libopencv_superres.so \ /usr/local/lib/libopencv_videoio.so \ /usr/local/lib/libopencv_video.so \ /usr/local/lib/libopencv_videostab.so # caffe INCLUDEPATH += ../include/caffe/include\ ../include/caffe/src \LIBS += /home/jinshan/caffe/build/lib/libcaffe.so LIBS +=../lib/libcaffe.so.1.0.0-rc3 LIBS += -L../lib LIBS += -lglog -lgflags -lprotobuf -lboost_system -lboost_thread -llmdb -lleveldb -lstdc++ -lcblas -latlasQMAKE_CXXFLAGS += -std=c++0x CONFIG += c++11

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12
• 13
• 14
• 15
• 16
• 17
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32
• 33
• 34
• 35
• 36
• 37
• 38
• 39
• 40
• 41
• 42
• 43
• 44
• 45
• 46
• 47
• 48
• 49
• 50
• 51
• 52
• 53
• 54
• 55

• lenet.h.創(chuàng)建一個lenet類,包含網(wǎng)絡(luò)的具體調(diào)用和數(shù)據(jù)預(yù)處理(核心代碼)?
  
  • Lenet(std::string model_file,std::string delay_file)://初始化網(wǎng)絡(luò),加載權(quán)重
  • void WrapInputLayer(const vector imgs, std::vector *input_channels); //多張圖片,h*w*3->3*h*w?
    void WrapInputLayer(const cv::Mat imgs, std::vector *input_channels); //單張圖片,h*w*3->3*h*w
  • int Predict(cv::Mat img); //返回預(yù)測結(jié)果
  • std::shared_ptr< Net>net; //定義網(wǎng)絡(luò)

#ifndef LENET_H #define LENET_H#define CPU_ONLY //CPU Mode#include <caffe/caffe.hpp> #include <opencv2/opencv.hpp> #include <string> #include <vector> #include <math.h> #include <QMessageBox> #include <qfile.h> #include <QTextStream> #include <QStatusBar> #include <stdio.h> #include <stdlib.h> #include <iostream> using namespace std; using namespace caffe;class Lenet { public:Lenet();~Lenet();Lenet(std::string model_file,std::string delay_file); //model_file->weight delay->file->netFilevoid WrapInputLayer(const vector<cv::Mat> imgs, std::vector<cv::Mat> *input_channels); //多張圖片,h*w*3->3*h*wvoid WrapInputLayer(const cv::Mat imgs, std::vector<cv::Mat> *input_channels); //單張圖片,h*w*3->3*h*wint Predict(cv::Mat img); //返回結(jié)果std::shared_ptr< Net<float>>net; //網(wǎng)絡(luò) }; #endif // LENET_H

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12
• 13
• 14
• 15
• 16
• 17
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32

• lenet.cpp

#include "lenet.h"Lenet::Lenet(){}Lenet::~Lenet(){}Lenet::Lenet(std::string model_file,std::string delay_file) {//CPU模式#ifdef CPU_ONLYCaffe::set_mode(Caffe::CPU);//GPU模式#elseCaffe::set_mode(Caffe::GPU);#endifnet.reset(new Net<float>(delay_file,TEST)); //加載網(wǎng)絡(luò)文件net->CopyTrainedLayersFrom(model_file); //加載權(quán)重文件 } //將w*h*3轉(zhuǎn)化位3*h*w void Lenet:: WrapInputLayer(const vector<cv::Mat> imgs, std::vector<cv::Mat> *input_channels) {Blob<float> *input_layer = net->input_blobs()[0]; //取數(shù)據(jù)層blobint width = input_layer->width(); //獲得blob的寬度int height = input_layer->height(); //獲得blob高度int num = input_layer->num(); //blob數(shù)量float *input_data = input_layer->mutable_cpu_data(); //指針方向指向數(shù)據(jù)blobfor (int j = 0; j < num; j++) { //input_data ->input_channels;//std::vector<cv::Mat> *input_channels;for (int k = 0; k < input_layer->channels(); ++k){cv::Mat channel(height, width, CV_32FC1, input_data);input_channels->push_back(channel);input_data += width * height;}cv::Mat img = imgs[j];cv::split(img, *input_channels);//賦值->給網(wǎng)絡(luò)賦值,分為三通道:r,g,binput_channels->clear();} }//單張圖片,h*w*3->3*h*wvoid Lenet::WrapInputLayer(const cv::Mat imgs, std::vector<cv::Mat> *input_channels){Blob<float> *input_layer = net->input_blobs()[0]; //取數(shù)據(jù)層blobint width = input_layer->width(); //獲得blob的寬度int height = input_layer->height(); //獲得blob高度float *input_data = input_layer->mutable_cpu_data(); //指針方向指向數(shù)據(jù)blobfor (int k = 0; k < input_layer->channels(); ++k){cv::Mat channel(height, width, CV_32FC1, input_data);input_channels->push_back(channel);input_data += width * height;}cv::split(imgs, *input_channels);//賦值->給網(wǎng)絡(luò)賦值,分為三通道:r,g,b}//預(yù)測結(jié)果int Lenet::Predict(cv::Mat img){std::vector<float> softMax;cv::Mat grayImg;if(img.channels()==3){cv::cvtColor(img, grayImg, cv::COLOR_BGR2GRAY);}//輸入必須位灰度圖cv::resize(grayImg,grayImg, cv::Size(28,28),cv::INTER_CUBIC); //輸入為28*28//cv::imwrite("w.jpg", grayImg);grayImg.convertTo(grayImg,CV_32FC1, 0.00392157); //歸一化 //rehape net 設(shè)置網(wǎng)絡(luò)輸入尺寸Blob<float> *input_layer = net->input_blobs()[0];input_layer->Reshape(1, 1,grayImg.rows, grayImg.cols);net->Reshape();//change channels 數(shù)據(jù)轉(zhuǎn)換std::vector<cv::Mat>channels;WrapInputLayer(grayImg,&channels);//forword net 前向傳播net->Forward();//get SOftMax label 取網(wǎng)絡(luò)最后一層SoftMaxBlob<float>* out_layer = net->output_blobs()[0];int count = out_layer->count();const float* start_feature = out_layer->cpu_data();const float* end_feature = out_layer->cpu_data() + count;softMax =std::vector<float>(start_feature, end_feature);//獲得最大值的下標index=預(yù)測數(shù)字float result = softMax[0];int index = 0;cout<< "sfds"<<softMax.size()<<endl;for(int i = 1; i < softMax.size();i++){if(result< softMax[i]){result = softMax[i];index = i;}cout<<i<<":"<<softMax[i];}cout<<"result:"<<index<<" ";return index;}

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12
• 13
• 14
• 15
• 16
• 17
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32
• 33
• 34
• 35
• 36
• 37
• 38
• 39
• 40
• 41
• 42
• 43
• 44
• 45
• 46
• 47
• 48
• 49
• 50
• 51
• 52
• 53
• 54
• 55
• 56
• 57
• 58
• 59
• 60
• 61
• 62
• 63
• 64
• 65
• 66
• 67
• 68
• 69
• 70
• 71
• 72
• 73
• 74
• 75
• 76
• 77
• 78
• 79
• 80
• 81
• 82
• 83
• 84
• 85
• 86
• 87
• 88
• 89
• 90
• 91
• 92
• 93
• 94
• 95
• 96
• 97
• 98
• 99
• 100
• 101
• 102
• 103
• 104
• 105
• 106
• 107

• mainwindow.cpp,網(wǎng)絡(luò)調(diào)用和初始化,選擇圖片進行處理得出預(yù)測

#include "mainwindow.h" #include "ui_mainwindow.h" #include "lenet.h" MainWindow::MainWindow(QWidget *parent) :QMainWindow(parent),ui(new Ui::MainWindow) {ui->setupUi(this);//設(shè)置窗口this->showMaximized(); //最大化ui->srcImage->setScaledContents(true); //圖片自適應(yīng)label//初始化網(wǎng)絡(luò)string model_file = "./lenet_iter_10000.caffemodel"; //權(quán)重文件string delay_file = "./lenet_delay.prototxt"; //網(wǎng)絡(luò)文件lenetCNN = Lenet(model_file,delay_file); //初始化 }MainWindow::~MainWindow() {delete ui; }void MainWindow::on_selectImage_clicked() //按鈕selectImage的槽函數(shù) {//獲得圖片路徑QString fileName = QFileDialog::getOpenFileName(this, "open image",QDir::currentPath(),"All files(*.jpg *.png *.bmp)");//std::cout<<fileName.toStdString();cv::Mat img; //定義圖片int resultNumber; //識別結(jié)果img = cv::imread(fileName.toStdString()); //讀取圖片resultNumber = lenetCNN.Predict(img); //預(yù)測ui->reseult->setText("number: " + QString::number(resultNumber)); //顯示// cout<<"number:"<<resultNumber;//圖片顯示到srcImage控件上cv::cvtColor(img, img, CV_BGR2RGB);QImage qImg = QImage((const unsigned char*)(img.data), img.cols, img.rows, img.cols * img.channels(), QImage::Format_RGB888);ui->srcImage->clear();ui->srcImage->setPixmap(QPixmap::fromImage(qImg)); }

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12
• 13
• 14
• 15
• 16
• 17
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32
• 33
• 34
• 35
• 36
• 37
• 38
• 39
• 40
• 41
• 42
• 43
• 44
• 45

• main.cpp 沒有改動

#include "mainwindow.h" #include <QApplication>int main(int argc, char *argv[]) {QApplication a(argc, argv);MainWindow w;w.show();return a.exec(); }

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12

• mainwindow.ui是界面文件,自行下載源碼查看

版權(quán)聲明：本文為博主原創(chuàng)文章，未經(jīng)博主允許不得轉(zhuǎn)載。 https://blog.csdn.net/jmu201521121021/article/details/79048220

總結(jié)

以上是生活随笔為你收集整理的深度学习入门篇(二)Lenet网络在caffe+QtCreator上部署应用的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。