ManualKDE: 交互式选点 x 核密度估计 x 热力图可视化 x 参数可调

使用说明：

输出结果存储在当面目录下的results文件夹（自动生成），命名为原始图片文件名+'_heat.png'

不用多次运行代码，需要用多少次就用多少次：

“动一下鼠标（选图片） 几下鼠标 （标点） 继续动一下鼠标（选图片）”

特点

交互式选点：在底图上单击增加点，再次单击去除点

核密度估计：基于选点估算核密度

热力图可视化：可调颜色图，经典配色方案：‘jet'，'viridis'，'coolwarm'

参数可调：（1）输出图像大小(比例） out_shape；（2）核密度（高斯核）估计带宽 bandwidth

（3）数据集所在目录

package友好：家常库

and GPT目前不能一条龙解决

代码如下：

import numpy as np
from sklearn.neighbors import KernelDensity
import matplotlib.pyplot as plt
import cv2
import tkinter as tk
from tkinter import filedialog
import os



def ManualControlKDE(bw=40, default_directory='./', outshape=1.0):

    def toggle_neighbors(matrix, y, x):
        # 切换当前点的状态
        matrix[y, x] = 1 - matrix[y, x]

        # 切换四个相邻点的状态
        neighbors = [(y-1, x), (y+1, x), (y, x-1), (y, x+1)]
        for neighbor_y, neighbor_x in neighbors:
            if 0 <= neighbor_y < matrix.shape[0] and 0 <= neighbor_x < matrix.shape[1]:
                matrix[neighbor_y, neighbor_x] = 1 - matrix[neighbor_y, neighbor_x]

    def onclick(event):
        if event.xdata is not None and event.ydata is not None:
            x = int(round(event.xdata))
            y = int(round(event.ydata))

            # 切换当前点及其相邻点的状态
            toggle_neighbors(matrix, y, x)

            # 更新图像
            update_plot()

    def update_plot():
        plt.clf()
        plt.imshow(img)
        plt.imshow(matrix, cmap='viridis', alpha=0.3)
        plt.title("Click on points (1 to select, 0 to deselect)")
        plt.draw()

    resume = True
    while resume:
        root = tk.Tk()
        root.withdraw()  # 隐藏主窗口

        # 打开文件选择对话框
        file_path = filedialog.askopenfilename(title="选择文件", initialdir=default_directory, filetypes=[("图像文件", "*.png;*.jpg;*.jpeg")])

        # 检查用户是否选择了文件
        if file_path:
            print("选择的文件路径:", file_path)
        else:
            print("未选择任何文件")
            resume = False

        root.destroy()

        image_path = file_path
        img = cv2.imread(image_path)
        img = cv2.cvtColor(cv2.resize(img, (int(outshape*img.shape[1]), int(outshape*img.shape[0]))), cv2.COLOR_BGR2RGB)

        # 初始化矩阵
        matrix = np.zeros((img.shape[0], img.shape[1]), dtype=int)

        # 通过交互式绘图选择点
        plt.imshow(img)
        plt.title("Click on points (1 to select, 0 to deselect)")
        cid = plt.gcf().canvas.mpl_connect('button_press_event', onclick)
        plt.show()

        HW = matrix
        data_points = np.argwhere(HW == 1)

        # 使用核密度估计
        kde = KernelDensity(bandwidth=bw, kernel='gaussian')
        kde.fit(data_points)

        # 生成网格坐标
        x, y = np.meshgrid(np.linspace(0, HW.shape[1] - 1, HW.shape[1]), np.linspace(0, HW.shape[0] - 1, HW.shape[0]))
        grid_points = np.c_[y.ravel(), x.ravel()]

        # 计算核密度估计
        log_density = kde.score_samples(grid_points)
        density_matrix = np.exp(log_density).reshape(x.shape)

        plt.figure()
        plt.imshow(img)

        # 叠加核密度图
        plt.imshow(density_matrix, cmap='jet', alpha=0.3, extent=(0, img.shape[1], img.shape[0], 0))

        # 关闭坐标轴
        plt.axis('off')

        # 设置空标题
        plt.title('')

        if not os.path.exists('./results/'):
            os.mkdir('./results/')

        # files = os.listdir('./results/')
        base = os.path.basename(file_path).split('.')[0]
        plt.savefig('./results/{}_heat.png'.format(base), bbox_inches='tight')

        # 显示图形
        plt.show()

default_directory = 'xxxxxx'  # 指定包含需要处理的图像的目录
bandwidth = 40  # 影响热图效果
outshape = 0.5  # 输出大小比例
ManualControlKDE(bandwidth, default_directory, outshape)