混淆矩阵计算遥感分类精度（总体精度、Kappa系数、用户精度、生产者精度等）

a = [200, 2  , 20 , 1  , 9  , 000, 000, 1  ]
b = [6  , 50 , 10 , 2  , 20 , 000, 4  , 6  ]
c = [12 , 6  , 300, 5  , 5  , 000, 000, 11 ]
d = [1  , 1  , 3  , 50 , 9  , 000, 000, 18 ]
e = [4  , 2  , 6  , 2  , 300, 7  , 1  , 22 ]
f = [000, 1  , 2  , 1  , 30 , 80 , 2  , 9  ]
g = [1  , 5  , 3  , 2  , 10 , 10 , 24 , 4  ]
h = [1  , 3  , 15 , 5  , 30 , 5  , 2  , 300]

Confusion_Matrix = [
    a, b, c, d, e, f, g, h
]

import copy

def OA(Confusion_Matrix):
    CM = Confusion_Matrix
    xia_list = []
    for i in range(len(CM)):
        xia_list.append(sum(CM[i]))
    xia = sum(xia_list)
    shang_list = []
    for i in range(len(CM)):
        shang_list.append(CM[i][i])
    shang = sum(shang_list)
    OA = shang / xia
    return OA

def KP(Overall_Accuracy, Confusion_Matrix):
    CM = Confusion_Matrix
    p0 = Overall_Accuracy
    pe_xia_list = []
    for i in range(len(CM)):
        pe_xia_list.append(sum(CM[i]))
    pe_xia = sum(pe_xia_list) ** 2
    CMT = list(map(list, zip(*CM))) # 转置
    pe_shang_list = []
    for i in range(len(CM)):
        pe_shang_list.append(sum(CM[i]) * sum(CMT[i]))
    pe_shang = sum(pe_shang_list)
    pe = pe_shang / pe_xia
    KP = (p0 - pe) / (1 - pe)
    return KP

def CE(Class, Confusion_Matrix):
    CM = Confusion_Matrix
    Class_list = CM[Class]
    Commission_list = copy.deepcopy(Class_list)
    del Commission_list[Class]
    CE = sum(Commission_list) / sum(Class_list)
    return CE

def OE(Class, Confusion_Matrix):
    CM = Confusion_Matrix
    CMT = list(map(list, zip(*CM))) # 转置
    Class_list = CMT[Class]
    Ommission_list = copy.deepcopy(Class_list)
    del Ommission_list[Class]
    OE = sum(Ommission_list) / sum(Class_list)
    return OE

# 总体精度
Overall_Accuracy = OA(Confusion_Matrix)
print("总体精度:",Overall_Accuracy)
# 卡帕系数
Kappa = KP(Overall_Accuracy, Confusion_Matrix)
print("卡帕系数:",Kappa)

# 针对某一类
# 错分误差
Commission_Error = CE(0, Confusion_Matrix) # 第一个参数位写第几类，0是第一类，1是第二类
print("错分误差:",Commission_Error)
# 漏分误差
Omission_Error = OE(0, Confusion_Matrix) # 第一个参数位写第几类，0是第一类，1是第二类
print("漏分误差:",Omission_Error)
# 用户精度
Users_Accuracy = 1 - Commission_Error
print("用户精度:",Users_Accuracy)
# 生产者精度/制图精度
Producers_Accuracy = 1 - Omission_Error
print("生产者精度/制图精度:",Producers_Accuracy)

用户精度 (User’s Accuracy):
假设分类结果的某类别的用户精度为 80%，则用户拿到分类结果后，10个像元中有8个是对的，2个是错的；

生产者精度 (Producer’s Accuracy):
假设分类结果的某类别的生产者精度为 80%，那么如果该类别有10个像素的话，则有2个没有分到该类别（丢失了）。

有一个很好的例子：
图像分类中混淆矩阵精度验证法中的几个指标说明 - GIS知乎-新一代GIS问答社区 (geoscene.cn)