Python Numpy数组阈值赋值异常求助：为何全部元素变为0.5？

我在Python中无法通过索引正确修改Numpy数组的元素，阈值处理逻辑失效。原本期望np.max(thrGradIntensity) == 1，但实际所有元素都变成了0.5，(thrGradIntensity==0.5).all()返回true，请问阈值处理逻辑为何失效？

问题代码 #

import numpy as np

maxVal = np.max(gradIntensity2) 
thrGradIntensity = gradIntensity2.copy()
highThr = maxVal / 5
lowThr = maxVal / 40
indHT = np.argwhere(gradIntensity2 >= highThr)
indLT = np.argwhere(gradIntensity2 <= lowThr)
ind1 = lowThr < gradIntensity2
ind2 = gradIntensity2 < highThr
ind3 = ind1 & ind2
ind = np.argwhere(ind3)
thrGradIntensity[indHT] = 1
thrGradIntensity[indLT] = 0
thrGradIntensity[ind] = 0.5

print(maxVal) #-----------------------------result= 425.9426808716647
print(highThr)#-----------------------------result= 85.18853617433294
print(lowThr) #------------------------------result= 10.648567021791617
print(np.max(thrGradIntensity)) #--------------result= 0.5
print((thrGradIntensity==0.5).all()) #---------------result= true

问题原因 #

• argwhere索引格式错误：np.argwhere()返回的是二维数组（每行对应一个元素的坐标），用这种二维索引直接赋值时，Numpy的广播机制会错误地将赋值操作应用到整个数组，而非目标元素。
• 赋值顺序叠加错误：即使索引格式正确，最后执行的thrGradIntensity[ind] = 0.5也可能覆盖之前的赋值，但核心问题还是索引格式导致的全局覆盖。

解决方法 #

方法1：使用布尔索引（推荐） #

直接用布尔条件作为索引，这是Numpy中修改元素的高效可靠方式：

import numpy as np

maxVal = np.max(gradIntensity2) 
thrGradIntensity = gradIntensity2.copy()
highThr = maxVal / 5
lowThr = maxVal / 40

# 直接通过布尔条件赋值
thrGradIntensity[gradIntensity2 >= highThr] = 1
thrGradIntensity[gradIntensity2 <= lowThr] = 0
thrGradIntensity[(gradIntensity2 > lowThr) & (gradIntensity2 < highThr)] = 0.5

# 验证结果
print(np.max(thrGradIntensity))  # 预期输出1
print((thrGradIntensity == 0.5).all())  # 预期输出False

方法2：修正argwhere索引格式 #

如果坚持使用argwhere，需要将返回的二维索引转换为一维tuple格式：

# 替换原赋值语句
thrGradIntensity[tuple(indHT.T)] = 1
thrGradIntensity[tuple(indLT.T)] = 0
thrGradIntensity[tuple(ind.T)] = 0.5

内容的提问来源于stack exchange，提问作者shahram