Page 9 - 南京医科大学自然版
P. 9
第44卷第9期 张瀚文,曹维娟,罗刚银,等. 基于改进型U⁃Net的变色油墨血浆判别模型[J].
2024年9月 南京医科大学学报(自然科学版),2024,44(9):1179-1189 ·1181 ·
Preparation of blood Add distilled High speed centrifu⁃
Extract Wash red Determination of
for scientific water and shake gation and suction of
whole blood blood cells hemoglobin cintent
research vigorously supernatant
图1 血红蛋白溶液制备流程
Figure 1 Preparation process of hemoglobin solution
A
Homoglobin: 0 g/L 0.050 g/L 0.100 g/L 0.150 g/L 0.200 g/L 0.250 g/L 0.275 g/L 0.300 g/L 0.325 g/L
B
Homoglobin: 0 g/L 0.050 g/L 0.100 g/L 0.150 g/L 0.200 g/L 0.250 g/L 0.275 g/L 0.300 g/L 0.325 g/L
A:Bagged plasma. B:Plasma extracted and placed in test tubes.
图2 血浆血红蛋白溶液比色板卡
Figure 2 Plasma hemoglobin solution colorimetric board
ö F1表示X轴与Y轴相互的连接操作,δ是控制模
ω æ 1
x
z = ç ÷ ∑ ( j,ω ) (3)
c
c
è W y ø 0 ≤ i ≤ H x 块大小的超参数,经映射函数后再将输出数据按X轴
将公式(2)和(3)的X轴和Y轴最大池化结果联 和Y轴方向分为两个特征矩阵,如式(5)、(6)所示。
h
h
合,随后载入 1×1 的卷积操作中获得区位注意力 g =σ(Fh (f )) (5)
ω
ω
Map。 g =σ(Fω (f )) (6)
ω
h
f=δ(F1 (z ,z )) (4) 上述二式分别经1×1卷积操作后,通过ReLU映
Input
Residual H×W×C
H×1×C X max pool Y max pool 1×W×C
Coneat+Conv2d 1×(W+H)×C/r
Batch Norm+Non⁃linenr
1×(W+H)×C/r
H×1×C Conv2d Conv2d H×W×C
Split
H×1×C ReLU ReLU H×W×C
Re⁃weight H×W×C
Output
图3 最大池化CA结构
Figure 3 Max pooling CA structure