# US 11,992,327 B2 · Methods for sweat volume detection based on sweat print imaging

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 11,992,327 B2
- Document ID: US11992327B2
- Status: Granted
- Kind: B2
- Application number: 18/451,822
- Filed: 2023-08-17
- Published: 2024-02-29
- Granted: 2024-05-28
- Earliest claimed: 2022-08-18
- Assignees: Chongqing Norsda Medical Device Co., Ltd.
- Applicants: Chongqing Norsda Medical Device Co., Ltd.
- Inventors: Xiaobo Xu; Ping Wang; Rong Li; Bin Fang; Jiacao Zhou
- Web page: https://patentgrep.com/patent/US11992327B2

## CPC

- A61B 5/14517: {for sweat}
- A61B 5/4266: {sweat secretion}
- A61B 5/0077: {Devices for viewing the surface of the body, e.g. camera, magnifying lens}
- A61B 5/1032: {Determining colour of tissue for diagnostic purposes}
- A61B 5/7264: {Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems}
- A61B 5/7267: {involving training the classification device}
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)

## Abstract

The present disclosure provides a method for sweat volume detection based on sweat print imaging, comprising: obtaining a raw data set, the raw data set including at least one sweat print image formed by dropping different volumes of sweat on a test paper under different environmental parameters; obtaining a color value extreme of the sweat print image and dividing the sweat print image into at least one region; calculating a sweat error based on a total sweat collection volume and a real sweat volume; obtaining a trained error prediction model; obtaining a to-be-tested sweat print image and current environmental parameters; inputting the current environmental parameters and a count of regions of the to-be tested sweat print image into the trained error prediction model to determine a sweat error of the to-be-tested sweat print image; calculating a sweat volume and a sweat rate based on the sweat error.

## Claims

### Claim 1

A method for sweat volume detection based on sweat print imaging, comprising:
data collecting: obtaining a raw data set, the raw data set including at least one sweat print image formed by dropping different volumes of sweat on a test paper under different environmental parameters, the environmental parameters including a temperature and a humidity;
image partitioning: identifying a color value of the at least one sweat print image, obtaining a color value extreme, and dividing the at least one sweat print image into a preset count of first regions according to the color value extreme;
model constructing:
calculating a total sweat collection volume of the at least one sweat print image and calculating a sweat error based on the total sweat collection volume and a real sweat volume, wherein the sweat error is obtained by:
obtaining a sweat collection volume of at least one first region and summing the sweat collection volume of the at least one first region to obtain the total sweat collection volume:
Q′i=(kx′1+b)a1+(kx′2+b)a2+, . . . ,(kx′p+b)ap, . . . ,+(kx′m+b)am,
wherein x′p denotes a median color value of a p-th region of the at least one sweat print image, ap denotes an area of the p-th region of the at least one sweat print image, k and b denote preset coefficients, (ky′p+b)ap denotes a sweat collection volume of the p-th region of the at least one sweat print image, Q′i denotes the total sweat collection volume, and m denotes a count of first regions; and
determining the sweat error based on a difference between the total sweat collection volume and the real sweat volume; and
constructing an error prediction model based on a Back-propagation (BP) neural network using the temperature, the humidity, and the count of first regions as an input of the error prediction model and the sweat error as an output of the error prediction model;
model training using the temperature, the humidity, and the count of first regions of the raw data set as the input of the error prediction model, and training parameters of the error prediction model to obtain a trained error prediction model; and
sweat volume predicting:
pasting a test paper to a to-be-tested part of a tester to obtain a to-be-tested sweat print image formed on the pasted test paper after a preset time period;
dividing the to-be-tested sweat print image into at least one region and calculating a total sweat collection volume of the to-be-tested sweat print image;
obtaining current environmental parameters, inputting the current environmental parameters and a count of second regions of the to-be-tested sweat print image into the trained error prediction model, and determining a sweat error of the to-be-tested sweat print image; and
calculating a sweat volume and a sweat rate of the to-be-tested part based on the sweat error of the to-be-tested sweat print image.

### Claim 2 (depends on 1)

The method of claim 1, wherein the data collecting is implemented by:
setting up at least one set of experimental environmental parameters and dropping different volumes of sweat on different test papers under each set of experimental environmental parameters; and
collecting sweat print images of different test papers to obtain the raw data set, wherein the test papers show different colors according to the different volumes, and the different colors correspond to different color values, and the raw data set is as follows:
S={s1(T1,H1,Q1,X1),s2(T2,H2,Q2,X2), . . . ,si(Ti,Hi,Qi,Xi), . . . ,sn(Tn,Hn,Qn,Xn)}
wherein S denotes the raw data set, si denotes raw data collected for an i-th time, n denotes a maximum count of sampling times, Ti denotes a temperature collected for the i-th time, Hi denotes a humidity collected for the i-th time, Qi denotes a real sweat volume collected for the i-th time, and Xi denotes the at least one sweat print image collected for the i-th time.

### Claim 3 (depends on 1)

The method of claim 1, wherein the count of first regions is determined by a following formula:
m
i
=
{
m
1
△
⁢
x
i
<
x
1
m
2
x
1
<
△
⁢
x
i
<
x
2
m
3
x
2
<
△
⁢
x
i
<
x
3
m
4
x
4
<
△
⁢
x
i
wherein mi denotes the count of first regions, x1, x2, x3, and x4 denote color value extreme thresholds, and Δxi denotes the color value extreme.

### Claim 4 (depends on 1)

The method of claim 1, wherein
the BP neural network includes one input layer, two hidden layers, and one output layer, wherein the input layer includes three inputs, each of the hidden layers includes two neurons, the output layer includes one output, the hidden layers adopt a Sigmoid function as an activation function, and the output layer adopts a Softmax function as an activation function.

### Claim 5 (depends on 1)

The method of claim 1, wherein the model training includes:
initializing a weight parameter and a bias parameter of the BP neural network, an initialized weight parameter being a random number between [−1, 1] obeying a normal distribution, and an initialized bias parameter being 0;
setting a loss function as:
Loss
=
1
2
⁢
∑
(
y
-
y
′
)
2
wherein y denotes a real sweat error of a training sample and y′ denotes a sweat error predicted by the BP neural network; and
setting a learning rate as α, updating the weight parameter and the bias parameter of the BP neural network using a gradient descent algorithm to obtain the trained error prediction model.

### Claim 6 (depends on 1)

The method of claim 1, the calculating the sweat volume and the sweat rate of the to-be-tested part based on the sweat error of the to-be-tested sweat print image includes:
summing the total sweat collection volume of the to-be-tested sweat print image and the sweat error of the to-be-tested sweat print image to obtain the sweat volume of the to-be-tested part and calculating the sweat rate of the to-be-tested part of the tester;
the sweat rate being obtained through a following formula:
R
=
Q
′
+
Δ
⁢
Q
′
a
wherein R denotes the sweat rate, a denotes an effective area of the test paper pasted to the to-be-tested part of the tester, Q′ denotes the total sweat collection volume of the to-be-tested sweat print image, and ΔQ′ denotes the sweat error of the to-be-tested sweat print image.

### Claim 7 (depends on 1)

The method of claim 1, wherein the environmental parameters further include a wind speed and the wind speed is obtained by an airflow sensor.

### Claim 8 (depends on 1)

The method of claim 1, wherein the image partitioning is further implemented by:
calculating an image gradient distribution of the at least one sweat print image, the image gradient distribution including an image gradient of each pixel point in the at least one sweat print image; and
dividing the at least one sweat print image into the at least one region using a preset division algorithm based on the color value extreme of the at least one sweat print image and the image gradient distribution, wherein the preset division algorithm includes an initialization stage, a region growing stage, and a region merging stage;
wherein the initialization stage includes:
determining a preset count of pixel points in the at least one sweat print image as seeds based on the color value extreme, and assigning a region label for each seed;
the region growing stage includes:
performing a plurality of rounds of iteration, at least one round of iteration including:
determining one or more seeds in a current round of iteration;
for each of the one or more seeds in the current round of iteration, calculating a first similarity between the seed and a neighboring pixel point of the seed, assigning a neighboring pixel point whose first similarity with the seed is greater than a first threshold with a region label corresponding to the seed, and determining the neighboring pixel point as a seed for a next round of iteration; the first similarity being determined based on a similarity of color values between two pixel points and a similarity of image gradients between the two pixel points; and
the region merging stage includes:
performing a plurality of rounds of iteration, at least one round of iteration including:
pairing to-be-merged regions in the at least one sweat print image, calculating a second similarity between to-be-merged regions in each pair, the second similarity being determined based on a first similarity between neighboring pixel points of the to-be-merged regions in the pair; and
merging the to-be-merged regions in the pair whose second similarity is greater than a second threshold; wherein the to-be-merged region refers to a connected region formed by pixel points with a same region label.

### Claim 9 (depends on 8)

The method of claim 8, wherein the second threshold is related to a difference in pore density of skin regions in which the to-be-merged regions in the pair are located.

### Claim 10 (depends on 1)

The method of claim 1, wherein each region corresponds to a set of preset coefficients, the preset coefficients of each region are determined based on a color value distribution feature of each region, and the color value distribution feature includes at least a color value histogram distribution and a color value dispersion degree of each pixel point within each region.

### Claim 11 (depends on 10)

The method of claim 10, wherein the preset coefficients of each region are further related to a pore density of a skin region in which each region is located.