Question

我正在尝试制作一个K最近邻分类器，以根据其色调颜色值识别对象。我有两个CSV文件填充了两个单独对象的这些颜色值，每个对象包含大约100个样本。它们都被归一化为包含40个整数，用于描述每行该对象的平均颜色。我使用K值10。

我的分类器总是返回标签“1”，即使我在我准备的训练数据相同的数据上使用它。当我查找函数find_nearest返回的最近点时，无论输入数据如何，我总能看到相同的10个点：1,1,1,1,1,1,1,1,1,1 。无论是输入数据还是学习数据，似乎都不对，但当我检查CvMat*个对象的内容时，它们就是应有的。

任何人都可以在下面的代码中看到我做错了什么吗？我通常使用c ++ API，但我发现找到k最近邻居的例子比较困难。我已在openCV和此blog

检查了示例

int dataArray[amountCategories_][amountSamples_][dataBandWidth_];
//Fill dataArray with data from csv files here (tested this and looks fine)

CvMat* trainData = cvCreateMat(amountCategories_ * amountSamples_, dataBandWidth_, CV_32FC1);
CvMat* trainClasses = cvCreateMat(amountCategories_ * amountSamples_, 1, CV_32FC1);

for (int categories = 0; categories < 2; categories++) {
    for (int samples = 0; samples < amountSamples_; samples++) {
        trainClasses->data.fl[samples] = categories;
        cout << samples << ": ";
        for (int datas = 0; datas < dataBandWidth_; datas++) {
          trainData->data.fl[samples * dataBandWidth_ + datas] = (float) dataArray[categories][samples][datas];
          cout << trainData->data.fl[samples * dataBandWidth_ + datas] << ", ";
        }
        cout << " and is type " << trainClasses->data.fl[samples] << endl;
    }
}

knn_.train(trainData, trainClasses);

识别对象：

vector<int> lineColors;
//fill lineColors with integers of the hue value
CvMat *playerRow = cvCreateMat(1, 40, CV_32FC1);
CvMat* nearests = cvCreateMat(1, kValue_, CV_32FC1);
cout << "Input vals: ";
for (int c = 0; c < dataBandWidth_; c++) {
    playerRow->data.fl[c] = (float) lineColors.at(c);
    cout << playerRow->data.fl[c] << ", ";
}
cout << " " << endl;
float outcome = knn_.find_nearest(playerRow, kValue_, 0, 0, nearests, 0);
cout << "Type: " << outcome << endl;

OpenCV Knearest无法正确分类

0 个答案: