当前位置：首页 > article >正文

根据YOLO数据集标签计算检测框内目标面积占比（YOLO7-10都适用）

article 2025/9/6 1:51:18

程序：

路径改成自己的，阈值可以修改也可以默认

#zhouzhichao
#25年5月17日
#计算时频图中信号面积占检测框面积的比值import os
import numpy as np
import pandas as pd
from PIL import Image# Define the path to the directory containing the label txt files
label_dir = r'D:\实验室\论文\论文-多信号参数估计\JCR3\Principle\微调原理\diff_snr\snr_dataset\label_txt'
img_dir = r'D:\实验室\论文\论文-多信号参数估计\JCR3\Principle\微调原理\diff_snr\snr_dataset\6'
binary_output_dir = r'D:\实验室\论文\论文-多信号参数估计\JCR3\Principle\像素面积比值\binary img'# Make sure the output directory exists
os.makedirs(binary_output_dir, exist_ok=True)# Initialize lists for storing the results for each class
ratios_0 = []
ratios_1 = []
ratios_2 = []
ratios_3 = []# Function to calculate the ratio of signal area to the bounding box area
def calculate_area_ratio(image_path, label_path, binary_output_dir):# Load the image and convert it to grayscaleimage = Image.open(image_path).convert('L')  # Convert to grayscale (L mode)image_array = np.array(image)# Read the label filewith open(label_path, 'r') as file:labels = file.readlines()for label in labels:# Parse the labellabel_data = label.strip().split()class_id = int(label_data[0])  # Class IDx_center = float(label_data[1])  # x center normalizedy_center = float(label_data[2])  # y center normalizedbox_width = float(label_data[3])  # box width normalizedbox_height = float(label_data[4])  # box height normalized# Convert the normalized values to pixel values based on image sizeimg_width, img_height = image.sizex_center_px = int(x_center * img_width)y_center_px = int(y_center * img_height)box_width_px = int(box_width * img_width)box_height_px = int(box_height * img_height)# Calculate the bounding box coordinatesx_min = max(x_center_px - box_width_px // 2, 0)y_min = max(y_center_px - box_height_px // 2, 0)x_max = min(x_center_px + box_width_px // 2, img_width)y_max = min(y_center_px + box_height_px // 2, img_height)# Extract the bounding box regionbbox_region = image_array[y_min:y_max, x_min:x_max]# Threshold the image (binary thresholding)binary_region = np.where(bbox_region > 128, 1, 0)  # 128 as threshold# Calculate the area of the signal inside the bounding box (sum of binary values)signal_area = np.sum(binary_region)# Calculate the area of the bounding boxbox_area = box_width_px * box_height_px# Save the binary image of the region to the output folderbinary_image = Image.fromarray(binary_region.astype(np.uint8) * 255)  # Convert to 0-255 scale for savingbinary_image.save(os.path.join(binary_output_dir, f'binary_{class_id}_{x_center_px}_{y_center_px}.png'))# Calculate the area ratioarea_ratio = signal_area / box_area if box_area > 0 else 0if class_id==0:ratios_0.append(area_ratio)if class_id==1:ratios_1.append(area_ratio)if class_id == 2:ratios_2.append(area_ratio)if class_id == 3:ratios_3.append(area_ratio)# Traverse all the label files and calculate area ratios
for label_file in os.listdir(label_dir):if label_file.endswith('.txt'):# Get corresponding image pathlabel_path = os.path.join(label_dir, label_file)image_name = label_file.replace('.txt', '.jpg')  # Assuming image is in .jpg formatimage_path = os.path.join(img_dir, image_name)# Calculate area ratios for the current label filecalculate_area_ratio(image_path, label_path, binary_output_dir)data = {'ratios_0': ratios_0,'ratios_1': ratios_1,'ratios_2': ratios_2,'ratios_3': ratios_3
}# 创建一个 DataFrame
df = pd.DataFrame(data)# 保存到 Excel 文件
file_path = r'D:\实验室\论文\论文-多信号参数估计\JCR3\Principle\像素面积比值\area_ratios.xlsx'
df.to_excel(file_path, index=False)
print()

程序依照的公式：