CN-121983345-A - Pet disease early warning method, device, equipment and storage medium

CN121983345ACN 121983345 ACN121983345 ACN 121983345ACN-121983345-A

Abstract

The method strictly distinguishes data processing from data level for acquired meteorological original data and biological original data, processes the meteorological original data and the biological original data respectively, and provides a corresponding data basis for a subsequent double-tower structure. The weather sequence tensor processing obtained by preprocessing the weather original data by using an independent weather tower, and the coded biological vector processing obtained by preprocessing the biological original data by using an independent biological tower are not mixed with each other, so that the corresponding weather features and biological features are separated at the structural level, the information of different sources is prevented from being mixed in the same network, the final prediction result is influenced, and the prediction accuracy of the model is improved.

Inventors

CHEN YAO
SONG JINZHAO
Yao xinwei

Assignees

浙江工业大学
中国科学院杭州医学研究所

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. A method for early warning of a pet disease, the method being applied to a trained pet disease early warning model, the pet disease early warning model comprising a meteorological tower, a biological tower, an exposure layer and a mission-head network, the method comprising: Acquiring meteorological original data and biological original data, wherein the biological original data is acquired by biological data acquisition terminals distributed at a plurality of diagnosis and treatment points and is uploaded to a server along with time stamps and geographic position identifiers; Preprocessing the meteorological original data and the biological original data respectively to correspondingly obtain a meteorological sequence tensor and a coded biological vector; Inputting the meteorological sequence tensor into a meteorological tower to obtain a corresponding meteorological feature vector, and inputting the encoded biological vector into a biological tower to obtain a corresponding biological feature vector; Inputting the meteorological sequence tensor into an exposure layer to obtain a daily exposure scalar sequence corresponding to each pathogen, and determining a corresponding exposure vector after all pathogens are combined according to the daily exposure scalar sequence of each pathogen; Determining corresponding intermediate gating features according to the meteorological feature vectors and the biological feature vectors, and determining corresponding gating fusion features according to the intermediate gating features; And after the gating fusion characteristic is spliced with the exposure vector, inputting the spliced gating fusion characteristic into a task head network to obtain a corresponding infection risk probability so as to early warn the pet diseases.
2. The pet disease pre-warning method according to claim 1, wherein the biological raw data includes age data, sex data, living environment data, season data and variety data, and the preprocessing is performed on the weather raw data and the biological raw data respectively, so as to obtain a weather sequence tensor and an encoded biological vector, and the method comprises: constructing weather sequence tensors with preset lengths according to the meteorological original data; normalizing the age data to obtain normalized age data; mapping the gender data, the living environment data, the season data and the variety data into integer indexes to obtain indexed multi-source data; And encoding the normalized age data and the indexed multi-source data to obtain an encoded biological vector.
3. The pet disease pre-warning method of claim 2, wherein the weather raw data comprises daily maximum air temperature data, daily minimum air temperature data, relative humidity data, diurnal temperature difference data, and air quality index data.
4. The pet disease pre-warning method according to claim 1, wherein the inputting the weather sequence tensor into a weather tower to obtain a corresponding weather feature vector, and inputting the encoded biological vector into a biological tower to obtain a corresponding biological feature vector, comprises: Inputting the weather sequence tensor with the preset length into a weather tower, and extracting time context information by using a single layer BiLSTM to obtain a daily weather feature vector; The daily weather feature vectors are respectively corresponding to weights based on the time attention layer, so that corresponding weather feature vectors are obtained; Inputting the coded biological vector into a biological tower, and processing by a multi-layer perceptron and a normalization layer to obtain a corresponding biological feature vector.
5. The pet disease pre-warning method according to claim 1, wherein inputting the weather sequence tensor into the exposure layer to obtain a daily exposure scalar sequence corresponding to each pathogen, and determining the exposure vector corresponding to each pathogen after combining according to the daily exposure scalar sequence of each pathogen, comprises: inputting the meteorological sequence tensors into an exposure layer to map to a daily exposure scalar for each pathogen; For each pathogen, carrying out time sequence modeling on a daily exposure scalar of the pathogen based on a corresponding LSTM unit, and weighting and accumulating the daily exposure scalar based on a gating mechanism to obtain an accumulated daily exposure sequence corresponding to each pathogen; the cumulative daily exposure sequences of all pathogens are combined to obtain the corresponding exposure vector.
6. The pet disease pre-warning method of claim 1, wherein the determining the corresponding intermediate gating feature from the meteorological feature vector and the biometric feature vector, and the determining the corresponding gating fusion feature from the intermediate gating feature, comprises: splicing and inputting the meteorological feature vector and the biological feature vector into a shared gating backbone network to obtain corresponding intermediate gating features; And linearly fusing the intermediate gating characteristics with the gating coefficients of all pathogens to obtain the gating fusion characteristics of all pathogens.
7. The pet disease pre-warning method of any one of claims 1-6, wherein the pet disease pre-warning model further comprises an interpretation interface, and wherein the method further comprises: the relative contributions of the meteorological towers, the biological towers, and the exposed layer in the mission head are evaluated for each pathogen's corresponding mission using the interpretation interface.
8. A pet disease pre-warning device, characterized in that the device is applied to a trained pet disease pre-warning model, the pet disease pre-warning model comprises a weather tower, a biological tower, an exposure layer and a task head network, the device comprises: the data acquisition module is used for acquiring meteorological original data and biological original data, wherein the biological original data are acquired by biological data acquisition terminals distributed at a plurality of diagnosis and treatment points and are uploaded to the server along with time stamps and geographic position identifiers; The preprocessing module is used for respectively preprocessing the meteorological original data and the biological original data to correspondingly obtain a meteorological sequence tensor and a coded biological vector; the feature extraction module is used for inputting the meteorological sequence tensor into a meteorological tower to obtain a corresponding meteorological feature vector, and inputting the encoded biological vector into the biological tower to obtain a corresponding biological feature vector; The exposure determining module is used for inputting the weather sequence tensor into the exposure layer to obtain a daily exposure scalar sequence corresponding to each pathogen, and determining the exposure vector corresponding to all the pathogens after combination according to the daily exposure scalar sequence of each pathogen; The feature fusion module is used for determining corresponding intermediate gating features according to the meteorological feature vectors and the biological feature vectors and determining corresponding gating fusion features according to the intermediate gating features; And the result generation module is used for inputting the spliced gating fusion characteristics and the exposure vectors into a task head network to obtain corresponding infection risk probability so as to early warn the pet diseases.
9. An electronic device, comprising: And a memory coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executable by the at least one processor to enable the at least one processor to perform the method of claim 1 7, A method for early warning of pet diseases.
10. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the pet disease pre-warning method according to any one of claims 1-7.

Description

Pet disease early warning method, device, equipment and storage medium Technical Field The disclosure relates to the technical field of computers, and in particular relates to a pet disease early warning method, device, equipment and storage medium. Background With the increasing demands of pet feeding, common pets such as cats and dogs have become important members of families, and the health status of the common pets is of great concern. The upper respiratory tract infection of the pets, which is caused by pathogens such as feline herpesvirus (Feline Herpesvirus, FHV), feline calicivirus (Feline Calicivirus, FCV) and influenza virus (Influenza A Virus, IAV), is a clinical high-rise disease, can cause symptoms such as cough, runny nose and the like of the pets, and can cause complications when serious, thereby bringing economic burden and emotional trouble to the owners of the pets. In order to avoid infection risk in advance, a pet disease early warning method based on environment and individual characteristics exists in the related technology, but the problem that the early warning practicability is insufficient and the disease prediction accuracy is low is commonly existed, and the method is difficult to support the fine pet health management. Disclosure of Invention The present disclosure provides a method, an apparatus, a device, and a storage medium for early warning of pet diseases, so as to at least solve the above technical problems in the prior art. In a first aspect of the present disclosure, there is provided a pet disease pre-warning method applied to a trained pet disease pre-warning model including a weather tower, a biological tower, an exposure layer, and a mission head network, the method comprising: Acquiring meteorological original data and biological original data, wherein the biological original data is acquired by biological data acquisition terminals distributed at a plurality of diagnosis and treatment points and is uploaded to a server along with time stamps and geographic position identifiers; Preprocessing the meteorological original data and the biological original data respectively to correspondingly obtain a meteorological sequence tensor and a coded biological vector; Inputting the meteorological sequence tensor into a meteorological tower to obtain a corresponding meteorological feature vector, and inputting the encoded biological vector into a biological tower to obtain a corresponding biological feature vector; Inputting the meteorological sequence tensor into an exposure layer to obtain a daily exposure scalar corresponding to each pathogen, and determining a corresponding exposure vector after all pathogens are combined according to the daily exposure scalar of each pathogen; Determining corresponding intermediate gating features according to the meteorological feature vectors and the biological feature vectors, and determining corresponding gating fusion features according to the intermediate gating features; And after the gating fusion characteristic is spliced with the exposure vector, inputting the spliced gating fusion characteristic into a task head network to obtain a corresponding infection risk probability so as to early warn the pet diseases. In an embodiment, the biological raw data includes age data, sex data, living environment data, season data and variety data, and correspondingly, the preprocessing is performed on the meteorological raw data and the biological raw data respectively, so as to obtain a meteorological sequence tensor and an encoded biological vector correspondingly, and the method includes: constructing weather sequence tensors with preset lengths according to the meteorological original data; normalizing the age data to obtain normalized age data; mapping the gender data, the living environment data, the season data and the variety data into integer indexes to obtain indexed multi-source data; And encoding the normalized age data and the indexed multi-source data to obtain an encoded biological vector. In one embodiment, the weather raw data includes daily maximum air temperature data, daily minimum air temperature data, relative humidity data, diurnal temperature difference data, and air quality index data. In an embodiment, the inputting the weather sequence tensor into a weather tower to obtain a corresponding weather feature vector, and inputting the encoded biological vector into a biological tower to obtain a corresponding biological feature vector, includes: Inputting the weather sequence tensor with the preset length into a weather tower, and extracting time context information by using a single layer BiLSTM to obtain a daily weather feature vector; The daily weather feature vectors are respectively corresponding to weights based on the time attention layer, so that corresponding weather feature vectors are obtained; Inputting the coded biological vector into a biological tower, and processing by a multi-layer perceptron and a norm