决策树完成图片分类任务

news2024/12/30 2:11:13

数据集要求：训练集和验证集（要求分好）

图片放置规则：一个总文件夹放类别名称的子文件夹其中子文件夹为存放同一类别图片

举个例子分类动物则总文件夹名称为动物子文件夹为猫狗猪猪。。。

其中猫的文件夹里面都是猫

给出代码：

import os
import cv2
import numpy as np
import logging
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score, confusion_matrix

# 设置日志
logging.basicConfig(filename='training_log.txt', level=logging.INFO, format='%(asctime)s - %(message)s')

# 读取图像数据和标签
def load_images_from_folder(folder):
    images = []
    labels = []
    label = 0
    for subdir in os.listdir(folder):
        subpath = os.path.join(folder, subdir)
        if os.path.isdir(subpath):
            for filename in os.listdir(subpath):
                if filename.endswith(".jpg"):
                    img_path = os.path.join(subpath, filename)
                    img = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)
                    img_normalized = cv2.resize(img, (256, 256))  # 归一化图像大小为256x256
                    images.append(img_normalized.flatten())
                    labels.append(label)
            label += 1
    return images, labels

# 主函数
def main():
    # train_folder = "YOUR_TRAIN_DATASET_FOLDER_PATH"  # 替换为你的训练集文件夹路径
    # test_folder = "YOUR_TEST_DATASET_FOLDER_PATH"    # 替换为你的测试集文件夹路径

    train_folder = "/Users/chen_dongdong/Desktop/宝钢项目/little_work/train"  # 替换为你的训练集文件夹路径
    test_folder = "/Users/chen_dongdong/Desktop/宝钢项目/little_work/val"    # 替换为你的测试集文件夹路径

    logging.info("Loading training data from %s", train_folder)
    X_train, y_train = load_images_from_folder(train_folder)
    logging.info("Loaded %d training samples", len(X_train))

    logging.info("Loading test data from %s", test_folder)
    X_test, y_test = load_images_from_folder(test_folder)
    logging.info("Loaded %d test samples", len(X_test))

    logging.info("Training DecisionTreeClassifier...")
    clf = DecisionTreeClassifier()
    clf.fit(X_train, y_train)
    logging.info("Training completed.")

    y_pred = clf.predict(X_test)
    accuracy = accuracy_score(y_test, y_pred)
    logging.info("Test Accuracy: %f", accuracy)

    cm = confusion_matrix(y_test, y_pred)
    cm_normalized = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
    logging.info("Normalized Confusion Matrix:")
    for row in cm_normalized:
        logging.info(" - ".join(map(lambda x: "{:.2f}".format(x), row)))

    # 打印特征重要性
    feature_importances = clf.feature_importances_
    top_features = np.argsort(feature_importances)[-10:]  # 打印最重要的10个特征
    logging.info("Top 10 important features:")
    for idx in top_features:
        logging.info("Feature %d: %f", idx, feature_importances[idx])

if __name__ == "__main__":
    main()

使用DecisionTreeClassifier的feature_importances_属性。这个属性会返回一个数组，其中每个值表示相应特征的重要性。值越大，特征越重要。

我们使用的是图像的灰度值作为特征，所以特征的数量会非常大（例如，对于256x256的图像，有65536个特征）。为了简化输出，我们可以只打印出最重要的特征。