C++ opencv4 ANN(神經網路) 範例

資料來源: chatgpt/gemini

chatgpt code

#include <opencv2/opencv.hpp>
#include <opencv2/ml/ml.hpp>
#include <iostream>
#include <vector>

using namespace cv;
using namespace cv::ml;
using namespace std;

int main() {
    // Create training data
    // Here, we are assuming 2 features for simplicity
    Mat trainData = (Mat_<float>(4, 2) << 0, 0, 
                                          0, 1,
                                          1, 0,
                                          1, 1);
    // The labels for each training sample
    Mat labels = (Mat_<int>(4, 1) << 0, 1, 1, 0);  // XOR problem

    // Create and configure the ANN
    Ptr<ANN_MLP> ann = ANN_MLP::create();
    
    // Set up the network architecture
    // Layer sizes: input (2), hidden layer (3), output (1)
    vector<int> layerSizes = {2, 3, 1};  // 2 inputs, 3 neurons in the hidden layer, 1 output
    ann->setLayerSizes(layerSizes);
    ann->setActivationFunction(ANN_MLP::SIGMOID_SYM, 1, 1);  // Sigmoid function
    ann->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER, 100, 1e-6));  // Stop after 100 iterations or a small change
    ann->setTrainMethod(ANN_MLP::BACKPROP, 0.1);  // Backpropagation with learning rate 0.1

    // Train the ANN
    ann->train(trainData, ROW_SAMPLE, labels);

    // Test the network
    Mat testData = (Mat_<float>(2, 2) << 0, 0, 
                                          1, 1);
    Mat results;
    ann->predict(testData, results);

    // Output results (Prediction)
    cout << "Prediction for (0,0): " << results.at<float>(0, 0) << endl;
    cout << "Prediction for (1,1): " << results.at<float>(1, 0) << endl;

    return 0;
}

gemini code

#include <iostream>
#include <vector>
#include <opencv2/opencv.hpp>
#include <opencv2/ml.hpp>

using namespace cv;
using namespace cv::ml;
using namespace std;

int main() {
    // 1. 準備訓練數據
    // 假設我們有 4 個二維數據點和它們的標籤 (0 或 1)
    Mat trainingData = (Mat_<float>(4, 2) << 0, 0,
                                             0, 1,
                                             1, 0,
                                             1, 1);
    Mat labels = (Mat_<int>(4, 1) << 0,
                                     1,
                                     1,
                                     0);

    // 2. 建立 ANN 模型
    // 設定網路的層數和每層的神經元數量
    // 這裡我們創建一個包含一個輸入層 (2 個神經元), 一個隱藏層 (3 個神經元), 和一個輸出層 (1 個神經元) 的網路
    Ptr<ANN_MLP> ann = ANN_MLP::create();
    Mat layerSizes = (Mat_<int>(1, 3) << 2, 3, 1);
    ann->setLayerSizes(layerSizes);
    ann->setActivationFunction(ANN_MLP::SIGMOID_SYM); // 使用 Sigmoid 激活函數
    ann->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 1000, 1e-6)); // 設定終止條件
    ann->setTrainMethod(ANN_MLP::BACKPROP); // 使用反向傳播演算法

    // 3. 訓練 ANN 模型
    cout << "開始訓練..." << endl;
    ann->train(trainingData, ml::ROW_SAMPLE, labels);
    cout << "訓練完成！" << endl;

    // 4. 使用訓練好的模型進行預測
    cout << "\n預測結果：" << endl;
    Mat testData = (Mat_<float>(2, 2) << 0.5, 0.5,
                                        1.5, 0.5);
    Mat predictedLabels;
    ann->predict(testData, predictedLabels);

    for (int i = 0; i < predictedLabels.rows; ++i) {
        cout << "輸入: (" << testData.at<float>(i, 0) << ", " << testData.at<float>(i, 1) << ") -> 預測標籤: " << predictedLabels.at<float>(i, 0) << endl;
    }

    // 5. (可選) 儲存訓練好的模型
    ann->save("trained_ann.xml");
    cout << "\n模型已儲存為 trained_ann.xml" << endl;

    // 6. (可選) 載入已儲存的模型
    Ptr<ANN_MLP> loadedAnn = Algorithm::load<ANN_MLP>("trained_ann.xml");
    cout << "\n載入的模型預測結果：" << endl;
    Mat loadedPredictedLabels;
    loadedAnn->predict(testData, loadedPredictedLabels);
    for (int i = 0; i < loadedPredictedLabels.rows; ++i) {
        cout << "輸入: (" << testData.at<float>(i, 0) << ", " << testData.at<float>(i, 1) << ") -> 預測標籤: " << loadedPredictedLabels.at<float>(i, 0) << endl;
    }

    return 0;
}