机器学习-KNN

一、相关知识点

二、利用KNN完成香蕉和苹果的识别

import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import cv2
import os
import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier


def preprocess(img):
    # image_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  # 灰度化
    # # 加入sobel算子用于提取边缘信息
    # # # 高斯滤波，去除噪声防止干扰边缘提取
    # gaussian = cv2.GaussianBlur(image_gray, (5, 5), sigmaX=2)  # sigmax为2是因为只有两种颜色
    # # 垂直梯度处理，(滤波)
    # image_ver = cv2.filter2D(gaussian, -1, kernel_ver)
    # # # 水平梯度处理，(滤波)
    # # image_hor = cv2.filter2D(gaussian, -1, kernel_hor)
    # image_ver = cv2.resize(image_ver,  target_size)
    # flattened_edge = image_ver.flatten()
    # # -----至此获取到了图像的边缘信息-----

    img = cv2.resize(img,  target_size)
    img_flat = img.flatten()  # 展平，否则knn无法读取
    # -----以上是图像的原本信息-----
    # combined_features = np.concatenate([flattened_edge, img_flat])
    # -----拼接，在不损失颜色信息的前提下同时加入边缘信息----
    return img_flat

# # sobel算子,用于提取边缘
# # 垂直梯度算子
# kernel_ver = np.array([[-1, 0, -1],
#                        [-2, 0, 2],
#                        [-1, 0, 1]])
# # 水平梯度算子
# kernel_hor = np.array([[-1, -2, -1],
#                        [0, 0, 0],
#                        [1, 2, 1]])


# 读取图片并预处理
images = []
target_size = (100, 100)
for file_name in os.listdir('../data/fruit/img'):
    img = cv2.imread(f'../data/fruit/img/{file_name}')
    img_flat = preprocess(img)
    images.append(img_flat)
print(len(images))
# 2. 转成 NumPy 数组 (n_samples, n_features)
X = np.array(images, dtype=np.float32)
# 3.标准化
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

# 4.读取标签
labels = []
for file_name in os.listdir('../data/fruit/label'):
    with open(f'../data/fruit/label/{file_name}', 'r', encoding='utf-8') as f:
        first_char = f.read(1)
        labels.append(first_char)
print(len(labels))

y = labels

# -----------------训练

# 分出训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(
    X,
    y,
    test_size=0.2,
    random_state=42
)

# 训练
for i in range(3, 15):
    estimator = KNeighborsClassifier(n_neighbors=i, weights='distance')
    estimator.fit(X_train, y_train)  # fit的X参数必须输入二维的

    # 评估
    y_pred = estimator.predict(X_test)

    # 准确率
    accuracy = estimator.score(X_test, y_test)
    print(accuracy)
    # 对比找出分类错误的样本
    misclassified_indices = np.where(y_pred != y_test)[0]
    print(misclassified_indices)

# 
estimator = KNeighborsClassifier(n_neighbors=5)
estimator.fit(X_train, y_train)  # fit的X参数必须输入二维的
sample = cv2.imread('F:\py_MachineLearning\MachineLearning\MachineLearning\img.png')
sample=preprocess(sample)
sample = sample.reshape(1, -1)
sample_pred = estimator.predict(sample)

print("测试：",sample_pred)