SEM图像分析实战：纤维形态的自动化测量

为什么自动化SEM分析

手动从SEM图像测量纤维直径和取向费时、主观、不可重复。Python脚本能在几秒内一致地处理每张图像。本指南使用scikit-image，专为科学图像分析设计的库。

import numpy as np
from skimage import io, filters, measure, morphology

第一步：加载和预处理

img = io.imread("sem_fiber_network.tif", as_gray=True)
img_norm = (img - img.min()) / (img.max() - img.min())

# 高斯去噪
from scipy.ndimage import gaussian_filter
img_denoised = gaussian_filter(img_norm, sigma=1.0)

第二步：阈值分割

methods = {"Otsu": filters.threshold_otsu(img_denoised),
           "Li": filters.threshold_li(img_denoised),
           "Yen": filters.threshold_yen(img_denoised)}
thresh = methods["Otsu"]  # Otsu通常最适合双峰纤维/背景图
binary = img_denoised > thresh
print(f"纤维覆盖率: {binary.mean()*100:.1f}%")

第三步：清理二值掩膜

cleaned = morphology.remove_small_objects(binary, min_size=50)
closed = morphology.binary_closing(cleaned, morphology.disk(2))
skeleton = morphology.skeletonize(closed)  # 用于取向分析

第四步：测量纤维属性

labels = measure.label(closed)
props = measure.regionprops_table(labels, properties=[
    "area", "eccentricity", "orientation",
    "major_axis_length", "minor_axis_length"
])
import pandas as pd
df_fibers = pd.DataFrame(props)
df_fibers["直径_px"] = df_fibers["minor_axis_length"]
df_fibers["长径比"] = df_fibers["major_axis_length"] / df_fibers["minor_axis_length"]
print(f"发现 {len(df_fibers)} 个纤维片段")
print(f"平均直径: {df_fibers['直径_px'].mean():.1f} +/- {df_fibers['直径_px'].std():.1f} px")

第五步：取向分析

from skimage.transform import radon
theta = np.linspace(0., 180., max(img.shape), endpoint=False)
sinogram = radon(skeleton, theta=theta)
dominant_angle = theta[np.argmax(sinogram.sum(axis=0))]
print(f"主取向: {dominant_angle:.1f} 度")

第六步：批量处理

import glob
def batch_analyze_sem(image_dir, scale_nm_per_px=10.0):
    results = []
    for filepath in glob.glob(f"{image_dir}/*.tif"):
        img = io.imread(filepath, as_gray=True)
        binary = (img-img.min())/(img.max()-img.min()) > filters.threshold_otsu(img)
        cleaned = morphology.remove_small_objects(binary, 50)
        props = measure.regionprops_table(measure.label(cleaned),
            properties=["minor_axis_length", "area"])
        df = pd.DataFrame(props)
        results.append({"file": filepath, "n_fibers": len(df),
            "mean_diameter_nm": df["minor_axis_length"].mean()*scale_nm_per_px})
    return pd.DataFrame(results)

常见坑

问题	解决
------	------
光照不均	用滚动球背景扣除
纤维粘连	用分水岭分割
过分割	增大min_size
标尺错误	始终验证像素-纳米标定

参考资料

• van der Walt, S. et al. (2014). scikit-image. PeerJ, 2, e453.
• GitHub: scikit-image/scikit-image