# US 12,112,545 B2 · Catenary icing detection method based on infrared imaging and meteorological monitoring

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

- Number: US 12,112,545 B2
- Document ID: US12112545B2
- Status: Granted
- Kind: B2
- Application number: 18/595,538
- Filed: 2024-03-05
- Published: 2024-09-12
- Granted: 2024-10-08
- Earliest claimed: 2023-03-07
- Assignees: Nanjing Joint Institute for Atmospheric Sciences
- Applicants: Nanjing Joint Institute for Atmospheric Sciences
- Inventors: Hong Wu; Yong Li; Duanyang Liu; Linyi Zhou; Shoupeng Zhu; Chengying Zhu; Yan Zeng; Hao Wu; Wei Qian
- Web page: https://patentgrep.com/patent/US12112545B2

## CPC

- G06V 20/52: Surveillance or monitoring of activities, e.g. for recognising suspicious objects (recognising microscopic objects G06V20/69)
- G06T 5/50: using two or more images, e.g. averaging or subtraction
- G06V 20/56: exterior to a vehicle by using sensors mounted on the vehicle
- H04N 23/11: for generating image signals from visible and infrared light wavelengths
- G06T 2207/20081: Training; Learning
- G06T 2207/20221: Image fusion; Image merging
- Y02A 90/10: Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION
- Y02A: TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE

## Abstract

The disclosure provides a catenary icing detection method based on infrared imaging and meteorological monitoring. In the disclosure, infrared illumination is used to detect the target, real-time images of the target are obtained through an infrared imaging sensor, and meteorological data are synchronously obtained through a meteorological monitoring unit, and normalized fusion processing is performed; the data are stored in a front-end edge computing unit, and through the deep learning model, the states of water accumulation, snow accumulation and icing of the target are automatically identified, and then sent to a background system through a transmission unit.

## Claims

### Claim 1

A catenary icing detection method based on infrared imaging and meteorological monitoring, comprising following steps:
S1, obtaining original data: obtaining an infrared image and a visible light image in real time of a target catenary through a camera device, and synchronously obtaining meteorological data in real time, where the meteorological data includes temperature, humidity, rainfall, snowfall and wind speed;
S2, constructing a sample data set: normalizing original infrared and visible image data obtained in the S1, and fusing with the meteorological data to establish the sample data set of a model;
S3, constructing a deep learning model of MobileNetV3, and training the deep learning model of MobileNetV3 through the sample data set obtained in the S2 to obtain a trained deep learning model; and
S4, evaluating and analyzing sample data through the deep learning model trained in the S3, so as to identify states of water accumulation, snow accumulation and icing on a surface of the target catenary;
wherein a process of constructing the sample data set in the S2 is as follows:
firstly, obtaining a surface temperature T0 of the catenary by Formula (1),
T
0
=
{
1
ε
[
1
τ
a
⁢
T
r
n
-
(
1
-
ε
)
⁢
T
u
n
-
(
1
τ
a
-
1
)
⁢
T
a
n
]
}
1
/
n
,
(
1
)
in the formula, ε represents a surface emissivity of the catenary, and here ε is 0.78 of copper; τa represents an atmospheric transmittance, with a value of 0.766; Tr represents a radiation imaging value, and here Tr is a pixel value corresponding to the infrared imaging image; a value of n is 5.33; Tu represents an equipment ambient temperature (K) and Ta represents an atmospheric temperature (K);
then, obtaining a Channel[0] value Rt of R-channel in single-channel infrared image data by synthesizing through Formula (2):
Rt=(T0+R0)×U (2),
in the formula, Rt represents the Channel[0] value; T0 represents the surface temperature of the catenary, R0 represents an R-channel value of visible light, and U represents a collected humidity value, with a range of 0.0-1.0; and
finally, fusing the Channel[0] value of R-channel in the synthesized infrared image data with a Channel[1] value of G-channel and a Channel[2] value of B-channel in the visible image as the sample data.

### Claim 2 (depends on 1)

The catenary icing detection method based on infrared imaging and meteorological monitoring according to claim 1, wherein the real-time infrared image has the same field of view as the visible image, and a resolution is 800*600.

### Claim 3 (depends on 1)

The catenary icing detection method based on infrared imaging and meteorological monitoring according to claim 1, wherein the emissivity 0.78 of surface oxidized copper is taken as the value c of the surface emissivity of the catenary.

### Claim 4 (depends on 1)

The catenary icing detection method based on infrared imaging and meteorological monitoring according to claim 1, wherein a process of training the deep learning model MobileNetV3 in the S3 is as follows:
(1) first, setting a gradient descent rate of model learning as 0.001, inputting sample data into the model in batches, and calculating a model output loss loss (x) of the i-th sample data in the batch by Formula (3):
loss(x)=−Σi=1Cyi log fi(x) (3),
in the formula, x represents an input sample, C represents a total number of categories, with a value of 4; yi represents a real label corresponding to the i-th category; fi (x) represents a corresponding model output value;
(2) then, calculating an average loss L of the sample data of the batch by Formula (4):
L
=
1
n
⁢
∑
i
n
⁢
❘
"\[LeftBracketingBar]"
loss
⁢
(
x
)
❘
"\[RightBracketingBar]"
,
(
4
)
in the formula, L represents the average loss of the batch; n represents a number of samples in the batch and loss (x) represents a loss value of a single sample;
(3) comparing the average loss L of the sample data of the batch with a model convergence threshold;
(4) when L is greater than the set model convergence threshold, updating model parameters by an Adam optimizer and repeating steps (1)-(3) for iterative training until L is less than the set model convergence threshold or the number of iterations reaches a set value; and
(5) finally, inputting a verification sample data set into the trained deep learning model MobileNetV3 for verification.

### Claim 5 (depends on 4)

The catenary icing detection method based on infrared imaging and meteorological monitoring according to claim 4, wherein the set value of the number of iterations in the step (4) of the S3 is 200.

### Claim 6 (depends on 1)

The catenary icing detection method based on infrared imaging and meteorological monitoring according to claim 1, wherein in the S1, when the real-time infrared image of the target catenary is obtained by the camera, the brightness of the target catenary is also improved by infrared illumination.