主页 / coding / 问题 / 79447890
Accepted
wosker4yan
Asked: 2025-02-18 18:27:59 +0800 CST

我怎样才能分离用 findpeaks 计算出的条纹

  • 772

我想分离我用 scipy findpeaks 计算的条纹（红色曲线），我该如何实现。我想将它们分离并存储在文本文件中。

在此处输入图片描述

import numpy as np
from scipy.signal import find_peaks
import matplotlib.pyplot as plt


X = np.load('X.npy')  
Y = np.load('Y.npy')  
P_new = np.load('P_new.npy')  

# Example data: Replace with your actual data
T = np.real(P_new)  # Simulating a 2D matrix

# Plot the original image
plt.figure()
plt.imshow(T, cmap='jet', aspect='auto')
plt.colorbar()
plt.title('Original Image')
plt.show()

# Peak detection parameters
min_peak_dist = 3   # Minimum distance between peaks
min_peak_h = 3e-5   # Minimum peak height

x_coords = []
y_coords = []

# Process all rows from top to bottom
for k in range(T.shape[0]):
      tex = T[k, :]
      peaks, _ = find_peaks(tex, distance=min_peak_dist, height=min_peak_h)

if peaks.size > 0:
      x_coords.extend(X[k, peaks])
      y_coords.extend(Y[k, peaks])

# Plot detected peaks
plt.figure()
plt.scatter(x_coords, y_coords, color='r', s=2)  # 's' controls marker size
plt.xlabel('X Coordinate')
plt.ylabel('Y Coordinate')
plt.title('Detected Fringes in Real-World Coordinates')
plt.colorbar()
plt.show()

绘图数据在这里

我想看到的只是像这样的独立边缘：在此处输入图片描述

以前我可以使用 cv2 方法轮廓来完成，_ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)，但这是从查找边缘开始，这对我来说不是很严格，就像查找实际数据的峰值一样。

有人能帮忙吗？谢谢

python

1 个回答

  1. Best Answer

    我使用 取得了合理的成功HDBSCAN。

    我首先运行程序find_peaks以查找沿y 轴（而不是沿 x 轴）的峰值 - 这些是黑线。然后我将图像剪切到蓝色方块内，并使用 HDBSCAN 对点进行聚类。最终的聚类结果为彩色。

    在此处输入图片描述

    要绘制特定聚类，您可以使用：

    view_cluster = 20

plt.figure(figsize=(3, 5))
plt.scatter(*clustered[view_cluster].T, marker='.', s=1)
plt.xlabel('x')
plt.ylabel('y')
plt.title(f'Fringe {view_cluster}')
plt.grid(linestyle=':')

    在此处输入图片描述

    解决方案

    加载数据并预处理：

    import numpy as np
from scipy.signal import find_peaks
import matplotlib.pyplot as plt
from matplotlib.patheffects import withTickedStroke

#
# Example data: Replace with your actual data
#
X_orig = np.load('X.npy')
Y_orig = np.load('Y.npy')

P_new = np.load('P_new.npy')
n_rows, n_cols = P_new.shape
T = np.real(P_new)  # Simulating a 2D matrix

#Resample X and Y to match resolution of P_new
# NB. I use the min/max values for simplicity
x_axis = np.linspace(X_orig.min(), X_orig.max(), num=n_cols)
y_axis = np.linspace(Y_orig.min(), Y_orig.max(), num=n_rows)

X_grid, Y_grid = np.meshgrid(x_axis, y_axis)

# Peak detection parameters
min_peak_dist = 3   # Minimum distance between peaks
min_peak_h = 3e-5   # Minimum peak height

coords = []

# Process all cols (x axis)
for col_ix in range(n_cols):
    peaks, _ = find_peaks(T[:, col_ix], distance=min_peak_dist, height=min_peak_h)

    if not len(peaks):
        continue

    x_coord = x_axis[col_ix]
    peaks_y = y_axis[peaks]
    coords.extend([(x_coord, peak_y) for peak_y in peaks_y])

coords = np.array(coords)
x_coords, y_coords = coords.T


#
#Clip unwanted regions
#
ignore_left_of = 5
ignore_below = 25
# ignore_shorter_than = 10

coords_clipped = coords[(x_coords > ignore_left_of) & (y_coords > ignore_below)]

    聚类并可视化：

    #
# HDBSCAN
#
from sklearn.cluster import HDBSCAN
coord_labels = HDBSCAN().fit_predict(coords_clipped)

clustered = [
    coords_clipped[coord_labels==cluster_id]
    for cluster_id in np.unique(coord_labels)
]

#
# Visualise
#
plt.figure()

#Original image
plt.matshow(
    T, extent=[x_axis.min(), x_axis.max(), y_axis.min(), y_axis.max()],
    cmap='Greys_r', origin='lower', vmax=5e-4, aspect='auto'
)
plt.gcf().set_size_inches(12, 6)

plt.colorbar(extend='both')

#Detected peaks
plt.scatter(x_coords, y_coords, color='black', marker='.', s=0.2, alpha=0.4)

# #Clip regions
plt.axvline(ignore_left_of, path_effects=[withTickedStroke(angle=135)])
plt.axhline(ignore_below, path_effects=[withTickedStroke(angle=-135)])

#Clusters
clustered = sorted(clustered, key=lambda members: members[:, 1].min())
for cluster_members in clustered:
    plt.plot(*cluster_members[::30].T, alpha=0.2, zorder=0, lw=8)

#Formatting
plt.xlabel('x')
plt.ylabel('y')
plt.show()
    • 2

相关问题