CN117312578B

CN117312578B - Construction method and system of non-genetic carrier spectrum

Info

Publication number: CN117312578B
Application number: CN202311595533.4A
Authority: CN
Inventors: 戴鹏飞; 周春姐
Original assignee: Shandong Kingfisher Intelligent Technology Co ltd; Yantai Cloud Software Co ltd
Current assignee: Shandong Kingfisher Intelligent Technology Co ltd; Yantai Cloud Software Co ltd
Priority date: 2023-11-28
Filing date: 2023-11-28
Publication date: 2024-02-23
Anticipated expiration: 2043-11-28
Also published as: CN117312578A

Abstract

The invention relates to the technical field of data processing, in particular to a method and a system for constructing a non-genetic carrier spectrum, which comprise the following steps: based on a multi-class public database and a geographic information system, the data mining technology is adopted to collect and clean the space-time data of the original non-genetic elements, normalization processing is carried out, and a purified non-genetic space-time data set is generated. According to the invention, a long-short-term memory network reveals a non-genetic time law, a space-time evolution model is optimized, a convolutional neural network and a transducer algorithm are used for deep learning of non-genetic audiovisual information, multi-modal semantic features are enriched, a systematic non-genetic knowledge map is constructed by a graph database and an ontology, structural knowledge support is provided for inheritance, reinforcement learning, community detection and key node recognition algorithm are used for optimizing inheritance path prediction, scientific decision support is provided, deep semantic relation of non-genetic elements is generated against the network, a non-genetic deep semantic association network is expanded, and powerful data and theoretical support are provided for inheritance and protection.

Description

Construction method and system of non-genetic carrier spectrum

Technical Field

The invention relates to the technical field of data processing, in particular to a method and a system for constructing a non-genetic carrier spectrum.

Background

The field of data processing technology includes a wide range of techniques and methods aimed at processing, analyzing, storing and visualizing various data types to extract useful information and knowledge. Developments in this field have focused on the collection, cleaning, analysis, visualization and application of data to address a variety of issues, from business decisions to scientific research, and the like.

The construction method of the non-genetic inheritance map is a solution scheme applying a data processing technology and is used for constructing the inheritance map of the non-material cultural heritage (non-heritage). The non-genetic carrier spectrum is a structured information representation for recording and carrying knowledge, skills, tradition and history of non-material cultural heritage. The main purpose is to preserve and inherit non-material cultural heritage so that offspring can learn and inherit. This includes various non-genetic elements such as traditional skills, music, dance, drama, language, food, ceremony, etc. The method creates a visual and easily understood non-genetic map through data acquisition, processing, visualization and sharing so as to show the relation and inheritance relationship among non-genetic elements, promote the inheritance and popularization of the non-genetic elements, and provide a useful tool for research, education and cultural propagation. By these means, non-genetic knowledge is preserved, not lost or forgotten so that offspring can learn and inherit this important cultural heritage.

In the existing construction method of the non-genetic carrier spectrum, the traditional method is rough in the disclosure of the transition rule of the non-genetic element, lacks of excavation of the depth of the time sequence, and is difficult to accurately simulate the non-genetic time-space evolution. In addition, the existing method always considers visual or audio information independently when processing multi-mode information, but cannot fully utilize the deep learning technology to extract comprehensive features, so that semantic understanding is incomplete. In knowledge management, the traditional non-genetic knowledge representation mostly adopts an unstructured mode, which is unfavorable for the retrieval and application of knowledge. In the aspect of inheritance path prediction, the support of an effective optimization algorithm is lacked, and a scientific inheritance strategy is difficult to form. The existing method is also superficial in exploring deep semantic relations among non-genetic elements, lacks deep learning and semantic association analysis, and is insufficient for supporting complex non-genetic cultural inheritance network construction.

Disclosure of Invention

The invention aims to solve the defects in the prior art, and provides a method and a system for constructing a non-genetic carrier spectrum.

In order to achieve the above purpose, the present invention adopts the following technical scheme: the construction method of the non-genetic carrier spectrum comprises the following steps:

S1: based on a multi-class public database and a geographic information system, collecting and cleaning the space-time data of original non-genetic elements by adopting a data mining technology, and performing normalization processing to generate a purified non-genetic space-time data set;

s2: based on the purified non-genetic space-time data set, a long-short-time memory network is adopted to reveal the change rule of non-genetic elements along with time flow, a space-time evolution model is constructed, and a non-genetic space-time transition trend model is generated;

s3: based on the non-genetic space-time transition trend model, deep learning is carried out on multi-modal information comprising vision and audio by adopting a convolutional neural network and a transducer algorithm, multiple semantic features of non-genetic elements are extracted and integrated into the model, and a non-genetic multi-modal semantic feature set is generated;

s4: based on the non-genetic multi-modal semantic feature set, carrying out mapping management on non-genetic elements and semantic features thereof by adopting a graph database and an ontology construction method to generate a non-genetic knowledge graph;

s5: based on the non-genetic knowledge map, predicting future inheritance conditions of non-genetic elements by adopting a reinforcement learning optimization algorithm, optimizing a transmission chain by utilizing a community detection and key node identification algorithm, and further generating optimized non-genetic inheritance path prediction;

S6: based on the optimized non-genetic carrier path prediction, introducing a generated countermeasure network, performing deep learning on a large carrier chain, mining deep semantic relations among non-genetic elements, and generating a non-genetic deep semantic association network;

the non-genetic carrying path prediction after purification specifically comprises a transition trend of the non-genetic element in a future time period, key nodes and carrying chains in a carrying process, and the non-genetic depth semantic association network specifically comprises semantic relations of similarity and relativity among the non-genetic elements.

As a further scheme of the invention, based on a multi-class public database and a geographic information system, the method adopts a data mining technology to collect and clean the space-time data of original non-genetic elements, and performs normalization processing, and the method specifically comprises the following steps of:

S101: based on the target public database, acquiring original non-genetic element information comprising space-time data by adopting a crawler technology, and extracting text to generate a non-genetic original information database;

s102: based on the non-genetic original information database, adopting a data preprocessing technology comprising null filling and abnormal value removal to clean data, and generating a cleaned non-genetic information database;

s103: based on the cleaned non-genetic information database, carrying out standardization processing on geographic position information by means of a geographic information system, converting geographic information with diversified formats into a unified format, and generating a standardized geographic position non-genetic information database;

s104: based on the non-genetic information database of the normalized geographic position, performing normalization processing of non-genetic element information by adopting a data normalization technology to generate a purified non-genetic space-time data set;

the crawler technology is specifically used for acquiring HTML source codes of a target public database and analyzing acquired information, the non-genetic original information database comprises names, categories, geographic positions and time information of multiple non-genetic elements, the geographic information system is specifically used for integrating the geographic information into the database, information inquiry and management of the geographic positions of the non-genetic elements are realized in retrieval, and the data standardization is specifically used for converting the information of the multiple non-genetic elements into uniform measurement standards or scales, so that the influence due to the difference of the measurement units is reduced.

As a further scheme of the invention, based on the purified non-genetic space-time data set, the method adopts a long-short-time memory network to reveal the change rule of non-genetic elements along with time flow, builds a space-time evolution model, and generates a non-genetic space-time transition trend model, which comprises the following steps:

s201: based on the purified non-genetic space-time data set, a data set dividing technology is adopted to divide a training set, a verification set and a test set, so as to generate a training sub-non-genetic data set, a verification sub-non-genetic data set and a test sub-non-genetic data set;

s202: based on the training sub non-genetic data set, a long-short-time memory network is imported to perform model training, and a non-genetic space-time model training stage model is generated;

s203: based on the non-genetic space-time model training stage model, introducing the non-genetic data set of the verifier to perform model verification, and generating a non-genetic space-time model verification stage model;

s204: based on the non-genetic space-time model verification stage model, introducing the test sub non-genetic data set, and generating a non-genetic space-time transition trend model through model test;

the data set division is specifically to divide the data set into a training set, a verification set and a test set for model training, optimization and testing, the long-short-term memory network is specifically a cyclic neural network for processing and predicting time sequence data or time sequence data, and the model verification mainly aims at optimizing parameters of a model and avoiding the phenomenon of over fitting or under fitting.

As a further scheme of the invention, based on the non-genetic space-time transition trend model, the multi-modal information comprising vision and audio is subjected to deep learning by adopting a convolutional neural network and a transducer algorithm, various semantic features of non-genetic elements are extracted and integrated into the model, and the step of generating a non-genetic multi-modal semantic feature set comprises the following steps:

s301: based on non-genetic visual data, performing feature coding by adopting a hierarchical feature extraction algorithm VGGNet, and performing feature selection to generate a visual feature set;

s302: based on non-genetic audio data, adopting Mel frequency cepstrum coefficient to perform voice coding, performing feature optimization, and generating an audio feature set by using the visual feature set;

s303: based on the audio feature set and the visual feature set, adopting a self-attention mechanism to perform feature fusion, performing time sequence feature learning, and generating a preliminary multi-mode semantic feature set;

s304: based on the preliminary multi-mode semantic feature set, a bi-directional encoder is adopted for representation, multi-mode data integration is carried out, feature mapping is carried out, and an optimized non-genetic multi-mode semantic feature set is generated;

the visual feature set is specifically a high-dimensional feature numerical representation comprising shapes, colors and textures, the audio feature set is specifically a feature combination comprising rhythms, tones and intensities, and the preliminary multi-modal semantic feature set is specifically a joint representation of fusion visual and audio features.

As a further scheme of the invention, based on the non-genetic multi-modal semantic feature set, a graph database and an ontology construction method are adopted to perform mapping management on non-genetic elements and semantic features thereof, and the step of generating a non-genetic knowledge graph comprises the following steps:

s401: based on the non-genetic multi-modal semantic feature set, adopting a graph database engine Neo4j to perform data modeling and structure optimization to generate graph structured data of non-genetic elements;

s402: based on the graph structured data, adopting an ontology modeling tool to define semantic relations, and carrying out ontology mapping to generate a non-genetic ontology model;

s403: based on the non-genetic ontology model, carrying out ontology and data fusion by adopting a knowledge graph fusion method, particularly entity alignment, and carrying out graph optimization to generate a non-genetic knowledge graph frame;

s404: based on the non-genetic knowledge graph frame, adopting a knowledge extraction and reasoning algorithm, calling SPARQL inquiry to carry out knowledge filling, and carrying out data verification to generate a complete non-genetic knowledge graph;

the graph structured data is specifically a set of nodes and edges of non-genetic elements and characteristics thereof, the non-genetic ontology model is specifically a formal description of semantic categories and attributes of the non-genetic elements, and the non-genetic knowledge graph frame is a hierarchical representation of structured non-genetic knowledge.

As a further scheme of the invention, based on the non-genetic knowledge map, a reinforcement learning optimization algorithm is adopted to predict the future inheritance situation of the non-genetic element, and a community detection and key node recognition algorithm is utilized to optimize the inheritance chain, so that the step of generating the optimized non-genetic inheritance path prediction is specifically as follows:

s501: based on the non-genetic knowledge graph, adopting a graph convolution network algorithm to perform feature extraction, and combining a deep Q network in reinforcement learning to generate a preliminary inheritance path prediction set;

s502: based on the preliminary inheritance path prediction set, adopting an evolution strategy algorithm to perform path selection optimization, and performing fine adjustment on a prediction result to generate an optimized inheritance path prediction set;

s503: based on the optimized inheritance path prediction set, community division is carried out by adopting a modularity maximization algorithm, and a community structure analysis set is generated by combining a label propagation algorithm;

s504: based on the community structure analysis set, performing key node identification by adopting a PageRank algorithm of field optimization, and performing chain optimization by combining with an influence maximization theory to generate an optimized non-genetic bearing path prediction;

the deep Q network specifically refers to an optimization algorithm combining deep learning and reinforcement learning, the evolution strategy algorithm specifically refers to a group-based optimization algorithm, the modularity maximizing algorithm specifically refers to an optimization algorithm for detecting a network community structure, nodes connected in communities are identified, and the field-optimized PageRank algorithm specifically refers to a traditional PageRank algorithm optimized by considering field information of the nodes.

As a further scheme of the invention, based on the optimized non-genetic carrier path prediction, an antagonism network is introduced and generated, a large carrier chain is subjected to deep learning, deep semantic relations among non-genetic elements are mined, and the step of generating a non-genetic deep semantic association network specifically comprises the following steps:

s601: based on the optimized non-genetic bearing path prediction, adopting a condition generation countermeasure network, carrying out characteristic expression learning, and carrying out sample generation to generate a characteristic learning set;

s602: based on the feature learning set, a stacked self-encoder is adopted to carry out depth feature coding and feature reconstruction, and a deep relation mining set is generated;

s603: based on the deep relation mining set, performing image and text feature extraction by adopting a convolutional neural network, and performing feature fusion to generate a deep feature set;

s604: based on the deep feature set, a long-time and short-time memory network is adopted to perform time sequence data modeling and relationship prediction, and a non-genetic deep semantic association network is generated;

the condition generation countermeasure network specifically comprises a condition variable introduced on the basis of generating a countermeasure network, the stacked self-encoders specifically comprise a depth network structure formed by stacking a plurality of groups of self-encoders and are used for learning high-order characteristic representation of data, and the convolutional neural network specifically refers to a deep learning algorithm and is used for analyzing visual images and extracting hierarchical characteristics in the images.

The system for constructing the non-genetic carrier spectrum is used for executing the method for constructing the non-genetic carrier spectrum, and comprises a non-genetic information processing module, a non-genetic space-time model constructing module, a non-genetic multi-modal feature extraction module, a non-genetic knowledge spectrum constructing module and a non-genetic deep learning and modeling module.

As a further scheme of the invention, the non-genetic information processing module adopts a crawler technology based on a disclosed target database to acquire original non-genetic element information comprising space-time data, and uses a data preprocessing technology to clean the original non-genetic element information to generate a purified non-genetic space-time data set;

the non-genetic space-time model building module adopts a data set dividing technology and a long-short-time memory network to perform model training and verification based on the generated purified non-genetic space-time data set to generate a non-genetic space-time transition trend model;

the non-genetic multi-modal feature extraction module performs feature extraction and optimization through a hierarchical feature extraction algorithm VGGNet and Mel frequency cepstrum coefficients based on non-genetic visual and audio data, performs feature fusion and data integration by using a self-attention mechanism and a bidirectional encoder representation, and generates an optimized non-genetic multi-modal semantic feature set;

The non-genetic knowledge graph construction module is based on an optimized non-genetic multi-modal semantic feature set, performs data modeling by using a graph database engine Neo4j, optimizes the constructed model by adopting an ontology modeling tool, and fuses and optimizes the ontology and the data by using a knowledge graph fusion method and a knowledge extraction algorithm with entity alignment to generate a complete non-genetic knowledge graph;

the non-genetic deep learning and modeling module performs feature extraction and inheritance path prediction by adopting a graph convolution network algorithm and combining a deep Q network in reinforcement learning according to a complete non-genetic knowledge map, performs path selection optimization and community division by adopting an evolution strategy algorithm and a modularity maximization algorithm, performs key node identification by using a PageRank algorithm of field optimization, and generates optimized non-genetic inheritance path prediction.

As a further scheme of the invention, the non-genetic information processing module comprises a data acquisition sub-module, a data preprocessing sub-module, a geographic position normalization sub-module and a data normalization sub-module;

the non-genetic space-time model building module comprises a data set dividing sub-module, a model training sub-module, a model verifying sub-module and a model testing sub-module;

The non-genetic multi-modal feature extraction module comprises a visual feature extraction sub-module, an audio feature extraction sub-module, a feature fusion sub-module and a multi-modal data integration sub-module;

the non-genetic knowledge map construction module comprises a data modeling sub-module, a structure optimization sub-module, a ontology data fusion sub-module and a knowledge extraction reasoning sub-module;

the non-genetic deep learning and modeling module comprises a feature extraction and prediction sub-module, a path optimization sub-module, a social region sub-module and a key node identification sub-module.

Compared with the prior art, the invention has the advantages and positive effects that:

according to the invention, the time law of non-genetic transition is revealed through the long-short-time memory network, so that the understanding capability of the model to the time sequence is enhanced, and the time-space evolution model is more accurate. The application of the convolutional neural network and the transducer algorithm carries out deep learning on visual and audio information of non-genetic elements, enriches multi-modal semantic feature extraction of the non-genetic elements, and improves the accuracy of semantic understanding. The graph database and the ontology construction method enable mapping management of non-genetic knowledge to be more systematic, and provide structural knowledge support for non-genetic elements and inheritance thereof. And the application of reinforcement learning, community detection and key node recognition algorithm optimizes the inheritance path prediction and provides more scientific decision support for non-inheritance paths. The generation of the countermeasure network is introduced, deep semantic relation among non-genetic elements is deeply mined, so that the non-genetic deep semantic association network is more comprehensive, and powerful data and theoretical support are provided for inheritance and protection of non-genetic culture.

Drawings

FIG. 1 is a schematic workflow diagram of the present invention;

FIG. 2 is a S1 refinement flowchart of the present invention;

FIG. 3 is a S2 refinement flowchart of the present invention;

FIG. 4 is a S3 refinement flowchart of the present invention;

FIG. 5 is a S4 refinement flowchart of the present invention;

FIG. 6 is a S5 refinement flowchart of the present invention;

FIG. 7 is a S6 refinement flowchart of the present invention;

FIG. 8 is a system flow diagram of the present invention;

FIG. 9 is a schematic diagram of a system framework of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Embodiment one:

referring to fig. 1, the present invention provides a technical solution: the construction method of the non-genetic carrier spectrum comprises the following steps:

s2: based on the purified non-genetic space-time data set, revealing the change rule of non-genetic elements along with time flow by adopting a long-short-time memory network, constructing a space-time evolution model, and generating a non-genetic space-time transition trend model;

s3: based on a non-genetic space-time transition trend model, deep learning is carried out on multi-modal information comprising vision and audio by adopting a convolutional neural network and a transducer algorithm, various semantic features of non-genetic elements are extracted and integrated into the model, and a non-genetic multi-modal semantic feature set is generated;

s4: based on a non-genetic multi-modal semantic feature set, carrying out mapping management on non-genetic elements and semantic features thereof by adopting a graph database and an ontology construction method to generate a non-genetic knowledge graph;

s5: based on the non-genetic knowledge graph, predicting future inheritance conditions of non-genetic elements by adopting a reinforcement learning optimization algorithm, optimizing a transmission chain by utilizing a community detection and key node identification algorithm, and further generating an optimized non-genetic inheritance path prediction;

the non-genetic carrying path prediction after purification comprises a non-genetic carrying path prediction specifically comprises a non-genetic carrying element carrying trend in a future time period, key nodes and carrying chains in a carrying process, and a non-genetic depth semantic association network specifically comprises similarity and correlation semantic relations among the non-genetic elements.

The data purification and normalization process effectively eliminates data noise in the S1 stage, improves data quality, and enables subsequent analysis to be more accurate and reliable. The space-time evolution model construction of S2 enhances the understanding of time sequences and the accuracy of the space-time model, and is helpful for deeper understanding of inheritance of non-genetic elements. S3 and S4 multi-mode semantic feature extraction and knowledge graph construction enrich semantic understanding and provide structural knowledge support for inheritance. And S5, predicting and optimizing the inheritance path to provide scientific decision support. And S6, understanding the non-genetic element relation by deep semantic mining extension provides comprehensive data and theoretical support for cultural inheritance.

Referring to fig. 2, based on a multi-class public database and a geographic information system, the data mining technology is adopted to collect and clean the spatiotemporal data of the original non-genetic elements, and normalization processing is performed, so that the steps of generating the purified non-genetic spatiotemporal data set are specifically as follows:

s104: based on a non-genetic information database of the normalized geographic position, carrying out normalization processing on non-genetic element information by adopting a data normalization technology, and generating a purified non-genetic space-time data set;

the crawler technology is specifically used for acquiring HTML source codes of a target public database and analyzing acquired information, the non-genetic original information database comprises names, categories, geographic positions and time information of multiple non-genetic elements, the geographic information system is specifically used for integrating the geographic information into the database, information inquiry and management of the geographic positions of the non-genetic elements are realized in retrieval, and data standardization is specifically used for converting the information of the multiple non-genetic elements into uniform measurement standards or scales, so that the influence due to the difference of measurement units is reduced.

In S101, original non-genetic element information including spatiotemporal data is acquired from a target public database using a crawler technique. By writing a crawler program that automatically accesses the database and parses the HTML source code to extract the required information. The extracted non-genetic element names, categories, geographical locations and time information are stored in a non-genetic original information database.

In S102, the non-genetic original information database is cleaned using a data preprocessing technique. This includes filling in empty values and removing outliers. For example, for missing geographic location information, a Geographic Information System (GIS) is used to supplement the missing data. And checking and correcting other errors or abnormal values, and ensuring the accuracy and consistency of the data.

In S103, the geographic position information is normalized by the geographic information system. Due to the format diversification of the geographic location information, such as longitude and latitude, address, etc., it is converted into a unified format for subsequent analysis and application. By using GIS tools and related algorithms such as coordinate conversion and address resolution.

In S104, the non-genetic element information is further normalized by using a data normalization technique. This includes converting the various non-genetic element information into a uniform metric or scale, reducing the impact due to unit of measure differences. For example, for time information, it is converted into a unified date format; for geographical location information, it is converted into a unified coordinate system. Through data normalization, consistency and comparability of the non-genetic space-time data set are ensured.

Referring to fig. 3, based on the purified non-genetic spatiotemporal data set, the method adopts a long-short-time memory network to reveal the change rule of non-genetic elements along with time flow, and the method comprises the following steps of:

s202: based on the training sub non-genetic data set, importing a long-short-time memory network to perform model training, and generating a non-genetic space-time model training stage model;

s203: training a phase model based on the non-genetic space-time model, introducing a non-genetic data set of a verifier, performing model verification, and generating a non-genetic space-time model verification phase model;

s204: based on a non-genetic space-time model verification stage model, a test sub non-genetic data set is introduced, and a non-genetic space-time transition trend model is generated through model test;

the data set division is specifically to divide the data set into a training set, a verification set and a test set, and is used for model training, optimization and testing, the long-short-time memory network is specifically a cyclic neural network and is used for processing and predicting time sequence data or time sequence data, and model verification mainly aims at optimizing parameters of a model and avoiding over-fitting or under-fitting phenomena.

In S201, the purified non-genetic spatiotemporal dataset is divided into a training set, a verification set and a test set. By using a data set partitioning technique, such as random partitioning or time-sequential partitioning. The divided training set is used for training the model, the verification set is used for optimizing parameters of the model, and the test set is used for evaluating the performance of the model.

In S202, the training sub non-genetic dataset is imported into a long short time memory network (LSTM) for model training. LSTM is a recurrent neural network that processes and predicts data with timing. In the training process, the data in the training set is used to adjust the parameters of the LSTM model so that the data can be better fitted.

In S203, a verifier non-genetic dataset is introduced for model verification. The model verification aims at optimizing parameters of the model and avoiding the phenomenon of over fitting or under fitting. And evaluating the performance of the model by comparing the difference between the predicted result and the actual result of the model on the verification set, and adjusting and optimizing the model according to the evaluation result.

In S204, the test sub non-genetic dataset is introduced into the validated non-genetic space-time model for model testing. And (3) evaluating the performance of the model on unknown data through the comparison of the predicted result and the actual result on the test set, and generating a final non-genetic space-time transition trend model. The model is used for analyzing and predicting the change rule of non-genetic elements along with time flow and providing reference basis for related decisions.

Referring to fig. 4, based on a non-genetic space-time transition trend model, the multi-modal information including vision and audio is deep learned by adopting a convolutional neural network and a transducer algorithm, multiple semantic features of non-genetic elements are extracted and integrated into the model, and the step of generating a non-genetic multi-modal semantic feature set is specifically as follows:

s302: based on non-genetic audio data, adopting Mel frequency cepstrum coefficient to perform speech coding, performing feature optimization, and generating an audio feature set by using a visual feature set;

s303: based on the audio feature set and the visual feature set, adopting a self-attention mechanism to perform feature fusion, and performing time sequence feature learning to generate a preliminary multi-mode semantic feature set;

In S301, feature encoding is performed using VGGNet (Visual Geometry Group Network). VGGNet is a classical convolutional neural network structure that is capable of extracting high-level features of an image through multi-layer convolution and pooling operations. In this step, non-residual visual data is input into the VGGNet network, resulting in a characteristic representation of each image. The visual feature set is generated by feature selection techniques such as Principal Component Analysis (PCA) or correlation analysis, selecting the most representative feature subset from the high-dimensional features.

# import VGGNet model

from keras.applications import VGG16

from keras.models import Model

from keras.layers import Flatten

# load Pre-trained VGGNet model, remove top layer (full connection layer)

base_model = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

# obtain output characteristics of VGGNet

visual_features = base_model.output

visual_features = Flatten()(visual_features)

Construction of a model for extracting visual characteristics

visual_model = Model(inputs=base_model.input, outputs=visual_features)

In S302, speech coding is performed using Mel frequency cepstral coefficients (Mel-frequency cepstral coefficients, MFCCs). MFCCs are a commonly used method of extracting audio features, which can effectively capture the spectral features of audio. In this step, the audio data is converted to MFCCs representation and feature optimized, e.g., mean normalized or normalized, to yield a stable audio feature set.

import librosa

import numpy as np

# loading audio data

audio_data, sr = librosa.load('audio.wav', sr=None)

# extraction of MFCCs features

mfccs = librosa.feature.mfcc(y=audio_data, sr=sr, n_mfcc=13)

# feature optimization (e.g., mean normalization)

mfccs_mean_normalized = np.mean(mfccs.T, axis=0)

S303: based on the audio feature set and the visual feature set, the Self-Attention mechanism (Self-Attention) is adopted to fuse the features in the feature fusion and time sequence feature learning. The self-attention mechanism can give different weights according to the relationship between features, thereby capturing the timing relationship between features better. This step generates a preliminary multimodal semantic feature set.

from keras.layers import Input, Dense, Dot, Activation

from keras.models import Model

# definition self-attention mechanism model

def self_attention(input_features):

attention = Dot(axes=[2, 2])([input_features, input_features])

attention = Activation('softmax')(attention)

output_features = Dot(axes=[2, 1])([attention, input_features])

return output_features

Construction of self-attention model #

input_audio_features = Input(shape=(time_steps, num_audio_features))

attended_audio_features = self_attention(input_audio_features)

In S304, the preliminary multi-mode semantic feature set is input into a bi-directional encoder to perform multi-mode data integration and feature mapping. The bidirectional encoder can simultaneously consider the context information of the input sequence, so that the association relation between the multi-modal features is better captured, and finally the optimized non-genetic multi-modal semantic feature set is obtained.

from keras.layers import Bidirectional, LSTM

# definition bidirectional encoder model

def bidirectional_encoder(input_features):

bidirectional_lstm = Bidirectional(LSTM(units=128, return_sequences=True))(input_features)

output_features = Dense(output_dim)(bidirectional_lstm)

return output_features

# build bi-directional encoder model

input_multimodal_features = Input(shape=(time_steps, num_visual_features + num_audio_features))

encoded_multimodal_features = bidirectional_encoder(input_multimodal_features)

Referring to fig. 5, based on a non-genetic multi-modal semantic feature set, a graph database and an ontology construction method are adopted to perform mapping management on non-genetic elements and semantic features thereof, and the step of generating a non-genetic knowledge map specifically includes:

s401: based on the non-genetic multi-modal semantic feature set, adopting a graph database engine Neo4j to perform data modeling and performing structure optimization to generate graph structured data of non-genetic elements;

s403: based on a non-genetic ontology model, adopting a knowledge graph fusion method, specifically entity alignment, to perform ontology and data fusion, and performing graph optimization to generate a non-genetic knowledge graph frame;

s404: based on a non-genetic knowledge graph frame, adopting a knowledge extraction and reasoning algorithm, calling SPARQL inquiry to carry out knowledge filling, and carrying out data verification to generate a complete non-genetic knowledge graph;

the graph structured data is specifically a collection of non-genetic elements and nodes and edges of the non-genetic elements, the non-genetic ontology model is specifically a formal description of semantic categories and attributes of the non-genetic elements, and the non-genetic knowledge graph framework is a hierarchical representation of structured non-genetic knowledge.

In S401, each element in the non-genetic multi-modal semantic feature set and its corresponding feature are represented as a node. This is accomplished by creating a node type to represent the non-genetic elements and assigning a unique identifier to each element. Relationship types are used to represent semantic relationships between different elements. For example, if two elements have similar characteristics or belong to the same category, a relationship type is created to represent such similarity or classification relationships. Each element's features corresponding to it and relationships to other elements are added to the graph. By creating nodes and relationships in the graph and connecting each other. And optimizing the structure of the graph, and improving the query efficiency and the storage space utilization rate. By merging duplicate nodes, deleting useless relationships, or adjusting labels of nodes and relationships.

In S402, semantic relationships between non-genetic elements are defined. This can be accomplished by using an ontology modeling tool to describe relationship types and properties between different elements. For example, a relationship type is defined to represent similarity between two elements and specify similarity scores for each other. And performing ontology mapping, and mapping the defined semantic relationship to nodes and relationships in a graph database. This is achieved by associating relationship types with relationships in the graph and mapping attributes to the nodes and to the attributes of the relationships. Ontological modeling tools are used to validate and correct defined semantic relationships and mapping results. By checking the correctness and consistency of the mapping. A non-genetic ontology model is generated that contains a formal description of the defined semantic relationships and the mapping results. This model is used for further knowledge graph construction and querying.

In S403, the non-genetic ontology model is fused with the graph structured data in the graph database using the entity alignment method. This is accomplished by mapping semantic relationships and attributes defined in the ontology model onto nodes and relationships in the graph. And optimizing the fused map, and improving the query efficiency and the storage space utilization rate. By merging duplicate nodes, deleting useless relationships, or adjusting labels of nodes and relationships. Generating a non-genetic knowledge graph framework, wherein the framework is specifically a hierarchical representation of the fused graph. This framework is used for further knowledge extraction and reasoning.

In S404, knowledge extraction algorithms are used to extract relevant knowledge and information from external data sources and add it to the non-genetic knowledge-graph. This is accomplished by invoking a SPARQL query that extracts the required information from the external data source according to the specified schema. Inference algorithms are used to infer and extend knowledge in the non-genetic knowledge graph. By applying logic rules and reasoning mechanisms, hidden associations and rules are discovered. And carrying out data verification on the filled knowledge to ensure the accuracy and consistency of the knowledge. By comparing the filled knowledge with an external data source or other knowledge graph for verification. And generating a complete non-genetic knowledge map which contains abundant non-genetic elements and information of semantic features thereof. This knowledge graph is used for further analysis and application.

Referring to fig. 6, based on a non-genetic knowledge graph, a reinforcement learning optimization algorithm is adopted to predict future inheritance conditions of non-genetic elements, and a community detection and key node recognition algorithm is utilized to optimize a transmission chain, so that the step of generating optimized non-genetic inheritance path prediction specifically comprises:

s501: based on a non-genetic knowledge graph, adopting a graph convolution network algorithm to perform feature extraction, and combining a deep Q network in reinforcement learning to generate a preliminary inheritance path prediction set;

s504: based on a community structure analysis set, performing key node identification by adopting a PageRank algorithm of field optimization, and performing chain optimization by combining an influence maximization theory to generate an optimized non-genetic bearing path prediction;

the deep Q network specifically refers to an optimization algorithm combining deep learning and reinforcement learning, the evolution strategy algorithm specifically refers to an optimization algorithm based on a group, the modularity maximization algorithm specifically refers to an optimization algorithm for detecting a network community structure, the nodes connected in the community are identified, and the field optimization PageRank algorithm specifically refers to a traditional PageRank algorithm optimized by considering field information of the nodes.

In S501, feature extraction is performed on the non-genetic knowledge graph using a graph convolution network algorithm. By taking non-genetic elements as nodes, inheritance relationships as edges, a graph structure is built, and graph rolling network models are used to learn representations of nodes and edges. A preliminary set of inheritance path predictions is generated in conjunction with a deep Q network in reinforcement learning. And optimizing the inheritance path prediction result of the non-genetic element through iterative training of the depth Q network.

In S502, a path selection optimization is performed on the preliminary inheritance path prediction set by using an evolutionary strategy algorithm, and a prediction result is finely tuned. The evolution strategy algorithm is a population-based optimization algorithm, simulates the biological evolution process, and searches for an optimal solution through continuous iterative selection and mutation operation. The inheritance path of the non-genetic element is regarded as a search problem, and the selection of the path is optimized through an evolutionary strategy algorithm, so that a more accurate inheritance path prediction result is obtained.

In S503, on the basis of optimizing the inheritance path prediction set, a modularity maximizing algorithm is adopted to perform community division, and a community structure analysis set is generated by combining with a tag propagation algorithm. The modularity maximization algorithm is an optimization algorithm for detecting the community structure of the network, and identifies the community structure formed by connected nodes in the network. By dividing the inheritance paths of the non-genetic elements into various communities, the inheritance modes and rules of the non-genetic elements are better understood.

In S504, based on the community structure analysis set, the key node identification is performed by adopting a PageRank algorithm of field optimization, and the chain optimization is performed by combining with an influence maximization theory, so that the optimized non-genetic bearing path prediction is generated. The field-optimized PageRank algorithm optimizes the traditional PageRank algorithm by considering the field information of the nodes, and more accurately identifies the nodes with important influence. By combining with the influence maximization theory, the non-genetic chain is further optimized, and the stability and the sustainability of the non-genetic chain are ensured.

Referring to fig. 7, based on the optimized non-genetic carrier path prediction, generating an antagonism network is introduced, deep learning is performed on a large carrier chain, deep semantic relations among non-genetic elements are mined, and the step of generating a non-genetic deep semantic association network specifically includes:

s602: based on the feature learning set, a stacked self-encoder is adopted to carry out depth feature encoding and feature reconstruction, and a deep relation mining set is generated;

s604: based on the deep feature set, a long-time and short-time memory network is adopted to perform time sequence data modeling, and relationship prediction is performed to generate a non-genetic deep semantic association network;

the condition generation countermeasure network specifically comprises a deep network structure formed by stacking a plurality of groups of self-encoders on the basis of generating a countermeasure network, wherein the deep network structure is used for learning high-order characteristic representation of data, and the convolutional neural network specifically refers to a deep learning algorithm and is used for analyzing visual images and extracting hierarchical characteristics in the images.

In S601, feature expression learning is performed on the induced condition generation countermeasure network based on the optimized non-genetic chain set. The condition generation countermeasure network introduces condition variables on the basis of generating the countermeasure network, and generates a characteristic representation with more distinction according to semantic information of non-genetic elements. And generating a feature learning set by iterating training conditions to generate an countermeasure network, wherein the feature learning set comprises deep semantic relations among non-genetic elements.

In S602, the feature learning set is depth feature coded and feature reconstructed using a stacked self-encoder. Stacked self-encoders are high-level feature representations of deep network structure learning data formed by stacking groups of self-encoders. Deep relation among non-genetic elements is further mined by extracting and reconstructing the features layer by layer, and a deep relation mining set is generated.

In S603, on the basis of the deep relation mining set, a convolutional neural network may be used to extract features of images and texts, and perform feature fusion. The convolutional neural network is a deep learning algorithm, and is particularly suitable for analyzing visual images and extracting hierarchical features in the images. By fusing the image and text features, a deep feature set is generated, wherein various types of feature information are integrated.

In S604, based on the deep feature set, the time series data modeling and the relation prediction are performed by using the long-short time memory network. The long-short time memory network is a deep learning model capable of processing time sequence data, and captures the relation and dependence on the time dimension. By training the long-short-term memory network, a non-genetic deep semantic association network is generated, and the network accurately predicts the relation and evolution trend among non-genetic elements.

Referring to fig. 8, a system for constructing a non-genetic carrier spectrum is used for executing the method for constructing a non-genetic carrier spectrum, and the system comprises a non-genetic information processing module, a non-genetic space-time model constructing module, a non-genetic multi-modal feature extraction module, a non-genetic knowledge graph constructing module, and a non-genetic deep learning and modeling module.

The non-genetic information processing module acquires original non-genetic element information comprising space-time data by adopting a crawler technology based on a disclosed target database, and cleans the original non-genetic element information by using a data preprocessing technology to generate a purified non-genetic space-time data set;

the non-genetic space-time model building module adopts a data set dividing technology and a long-short-time memory network to carry out model training and verification based on the generated purified non-genetic space-time data set, and generates a non-genetic space-time transition trend model;

The non-genetic multi-modal feature extraction module performs feature extraction and optimization through a hierarchical feature extraction algorithm VGGNet and Mel frequency cepstrum coefficients based on non-genetic visual and audio data, performs feature fusion and data integration by using a self-attention mechanism and a bi-directional encoder representation, and generates an optimized non-genetic multi-modal semantic feature set;

and the non-genetic deep learning and modeling module performs feature extraction and inheritance path prediction by adopting a graph rolling network algorithm and combining a deep Q network in reinforcement learning according to a complete non-genetic knowledge map, performs path selection optimization and community division by adopting an evolution strategy algorithm and a modularity maximizing algorithm, performs key node identification by utilizing a field-optimized PageRank algorithm, and generates optimized non-genetic inheritance path prediction.

The method comprises the steps of obtaining original non-genetic element information comprising space-time data from a disclosed target database through a crawler technology, and cleaning by using a data preprocessing technology to generate a purified non-genetic space-time data set. The accuracy and consistency of the data are ensured, and a reliable basis is provided for subsequent analysis and application. Based on the generated non-genetic space-time data set after purification, a data set dividing technology and a long-short-time memory network are adopted to carry out model training and verification, and a non-genetic space-time transition trend model is generated. Predicting and analyzing the inheritance path and development trend of the non-genetic element, and providing scientific basis for non-genetic protection and inheritance. Based on non-genetic visual and audio data, feature extraction and optimization are carried out through a hierarchical feature extraction algorithm VGGNet and Mel frequency cepstrum coefficients, feature fusion and data integration are carried out through a self-attention mechanism and a bi-directional encoder representation, and an optimized non-genetic multi-modal semantic feature set is generated. The visual and audio information of non-genetic elements are fully utilized, and the accuracy and the richness of feature expression are improved. Based on the optimized non-genetic multi-modal semantic feature set, performing data modeling by using a graph database engine Neo4j, optimizing the constructed model by using an ontology modeling tool, and fusing and optimizing the ontology and the data by using a knowledge graph fusion method and a knowledge extraction algorithm with entity alignment to generate a complete non-genetic knowledge graph. The relation and the relation among the non-genetic elements are displayed in a graphical mode, so that the user can conveniently understand and inquire. And according to the complete non-genetic knowledge graph, carrying out feature extraction and inheritance path prediction by adopting a graph rolling network algorithm and combining a deep Q network in reinforcement learning, carrying out path selection optimization and community division by adopting an evolution strategy algorithm and a modularity maximizing algorithm, and carrying out key node identification by utilizing a PageRank algorithm of field optimization to generate the optimized non-genetic inheritance path prediction. The inheritance path of the non-genetic element is deeply excavated and analyzed, and a more comprehensive and accurate reference is provided for non-genetic protection and inheritance.

Referring to fig. 9, the non-genetic information processing module includes a data acquisition sub-module, a data preprocessing sub-module, a geographic position normalization sub-module, and a data normalization sub-module;

the non-genetic space-time model building module comprises a data set dividing sub-module, a model training sub-module, a model verification sub-module and a model test sub-module;

In the non-genetic carrier spectrum construction system, a non-genetic information processing module is responsible for acquiring original non-genetic element information from a disclosed target database, and acquiring a data set comprising space-time data through a crawler technology. The data is cleaned and normalized, including noise removal, missing values processing, and geographic location information normalization.

In the non-genetic space-time model building module, model training is performed based on the cleaned data set. The data set is divided into a training set and a testing set through a data set dividing sub-module, a long-short-time memory network is used for model training, and the time-space transition trend of non-genetic elements is captured. The model verification sub-module and the model test sub-module are used for evaluating the performance and generalization capability of the model and ensuring the reliability of the model.

In the non-genetic multi-modal feature extraction module, the visual feature extraction submodule encodes non-genetic visual data by utilizing VGGNet and other algorithms, and the audio feature extraction submodule encodes audio data by using Mel frequency cepstrum coefficients. The feature fusion sub-module fuses the visual and audio features to form a multi-mode semantic feature set and captures rich information of non-genetic elements.

In the non-genetic knowledge graph construction module, a graph database engine Neo4j is utilized for data modeling, and the data modeling comprises a structure optimization sub-module, an ontology data fusion sub-module and a knowledge extraction reasoning sub-module. The method ensures the establishment of a knowledge graph, wherein the data structures of nodes and relations are optimized and fused, and the content and semantic consistency of the graph are enriched.

And in the non-genetic deep learning and modeling module, the graph rolling network and the deep Q network are comprehensively utilized to perform feature extraction and inheritance path prediction. The path optimization sub-module optimizes the transmission path through an evolution strategy algorithm and a modularity maximizing algorithm. The community dividing submodule divides non-genetic elements into various communities based on the topological structure of the atlas, and the key node identification submodule identifies key nodes in the atlas by using the PageRank algorithm optimized in the field, so that an optimized non-genetic chain-supporting set is formed.

The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.

Claims

1. The construction method of the non-genetic carrier spectrum is characterized by comprising the following steps:

based on a multi-class public database and a geographic information system, collecting and cleaning the space-time data of original non-genetic elements by adopting a data mining technology, and performing normalization processing to generate a purified non-genetic space-time data set;

based on the purified non-genetic space-time data set, a long-short-time memory network is adopted to reveal the change rule of non-genetic elements along with time flow, a space-time evolution model is constructed, and a non-genetic space-time transition trend model is generated;

based on the non-genetic space-time transition trend model, deep learning is carried out on multi-modal information comprising vision and audio by adopting a convolutional neural network and a transducer algorithm, multiple semantic features of non-genetic elements are extracted and integrated into the model, and a non-genetic multi-modal semantic feature set is generated;

Based on the non-genetic multi-modal semantic feature set, carrying out mapping management on non-genetic elements and semantic features thereof by adopting a graph database and an ontology construction method to generate a non-genetic knowledge graph;

based on the non-genetic knowledge map, predicting future inheritance conditions of non-genetic elements by adopting a reinforcement learning optimization algorithm, optimizing a transmission chain by utilizing a community detection and key node identification algorithm, and further generating optimized non-genetic inheritance path prediction;

based on the optimized non-genetic carrier path prediction, introducing a generated countermeasure network, performing deep learning on a large carrier chain, mining deep semantic relations among non-genetic elements, and generating a non-genetic deep semantic association network;

based on the purified non-genetic space-time data set, the long-short-time memory network is adopted to reveal the change rule of non-genetic elements along with time flow, the space-time evolution model is constructed, and the non-genetic space-time transition trend model is generated specifically by the following steps:

based on the purified non-genetic space-time data set, a data set dividing technology is adopted to divide a training set, a verification set and a test set, so as to generate a training sub-non-genetic data set, a verification sub-non-genetic data set and a test sub-non-genetic data set;

Based on the training sub non-genetic data set, a long-short-time memory network is imported to perform model training, and a non-genetic space-time model training stage model is generated;

based on the non-genetic space-time model training stage model, introducing the non-genetic data set of the verifier to perform model verification, and generating a non-genetic space-time model verification stage model;

based on the non-genetic space-time model verification stage model, introducing the test sub non-genetic data set, and generating a non-genetic space-time transition trend model through model test;

the data set division is specifically to divide the data set into a training set, a verification set and a test set for model training, optimization and testing, the long-short-time memory network is specifically a cyclic neural network for processing and predicting time sequence data or data with time sequence property, and the model verification mainly aims at optimizing parameters of a model and avoiding the phenomenon of over fitting or under fitting;

based on the non-genetic space-time transition trend model, the convolutional neural network and the transducer algorithm are adopted to carry out deep learning on the multi-modal information comprising the vision and the audio, multiple semantic features of the non-genetic elements are extracted and integrated into the model, and the step of generating the non-genetic multi-modal semantic feature set comprises the following steps:

Based on non-genetic visual data, performing feature coding by adopting a hierarchical feature extraction algorithm VGGNet, and performing feature selection to generate a visual feature set;

based on non-genetic audio data, adopting Mel frequency cepstrum coefficient to perform voice coding, performing feature optimization, and generating an audio feature set by using the visual feature set;

based on the audio feature set and the visual feature set, adopting a self-attention mechanism to perform feature fusion, performing time sequence feature learning, and generating a preliminary multi-mode semantic feature set;

based on the preliminary multi-mode semantic feature set, a bi-directional encoder is adopted for representation, multi-mode data integration is carried out, feature mapping is carried out, and an optimized non-genetic multi-mode semantic feature set is generated;

the visual feature set is specifically a high-dimensional feature numerical representation comprising shapes, colors and textures, the audio feature set is specifically a feature combination comprising rhythms, tones and intensities, and the preliminary multi-modal semantic feature set is specifically a joint representation of fusion visual and audio features;

2. The method for constructing a non-genetic carrier spectrum according to claim 1, wherein the step of collecting and cleaning the spatiotemporal data of the original non-genetic element and performing normalization processing based on a plurality of types of public databases and geographic information systems to generate the purified non-genetic spatiotemporal data set comprises the following steps:

based on the target public database, acquiring original non-genetic element information comprising space-time data by adopting a crawler technology, and extracting text to generate a non-genetic original information database;

based on the non-genetic original information database, adopting a data preprocessing technology comprising null filling and abnormal value removal to clean data, and generating a cleaned non-genetic information database;

based on the cleaned non-genetic information database, carrying out standardization processing on geographic position information by means of a geographic information system, converting geographic information with diversified formats into a unified format, and generating a standardized geographic position non-genetic information database;

based on the non-genetic information database of the normalized geographic position, performing normalization processing of non-genetic element information by adopting a data normalization technology to generate a purified non-genetic space-time data set;

3. The method for constructing a non-genetic carrier spectrum according to claim 1, wherein the step of generating the non-genetic knowledge spectrum by mapping and managing the non-genetic elements and the semantic features thereof by using a graph database and an ontology construction method based on the non-genetic multi-modal semantic feature set comprises the following steps:

based on the non-genetic multi-modal semantic feature set, adopting a graph database engine Neo4j to perform data modeling and structure optimization to generate graph structured data of non-genetic elements;

based on the graph structured data, adopting an ontology modeling tool to define semantic relations, and carrying out ontology mapping to generate a non-genetic ontology model;

Based on the non-genetic ontology model, carrying out ontology and data fusion by adopting a knowledge graph fusion method, particularly entity alignment, and carrying out graph optimization to generate a non-genetic knowledge graph frame;

based on the non-genetic knowledge graph frame, adopting a knowledge extraction and reasoning algorithm, calling SPARQL inquiry to carry out knowledge filling, and carrying out data verification to generate a complete non-genetic knowledge graph;

4. The method for constructing a non-genetic carrier spectrum according to claim 1, wherein the step of predicting future carrier conditions of non-genetic elements by using a reinforcement learning optimization algorithm based on the non-genetic knowledge spectrum and optimizing a transmission chain by using a community detection and key node recognition algorithm to generate an optimized non-genetic carrier path prediction is specifically as follows:

based on the non-genetic knowledge graph, adopting a graph convolution network algorithm to perform feature extraction, and combining a deep Q network in reinforcement learning to generate a preliminary inheritance path prediction set;

Based on the preliminary inheritance path prediction set, adopting an evolution strategy algorithm to perform path selection optimization, and performing fine adjustment on a prediction result to generate an optimized inheritance path prediction set;

based on the optimized inheritance path prediction set, community division is carried out by adopting a modularity maximization algorithm, and a community structure analysis set is generated by combining a label propagation algorithm;

based on the community structure analysis set, performing key node identification by adopting a PageRank algorithm of field optimization, and performing chain optimization by combining with an influence maximization theory to generate an optimized non-genetic bearing path prediction;

5. The method for constructing a non-genetic carrier spectrum according to claim 1, wherein the generation of an countermeasure network is introduced based on the optimized non-genetic carrier path prediction, deep learning is carried out on a large inheritance chain, deep semantic relations among non-genetic elements are mined, and the step of generating a non-genetic deep semantic association network is specifically as follows:

Based on the optimized non-genetic bearing path prediction, adopting a condition generation countermeasure network, carrying out characteristic expression learning, and carrying out sample generation to generate a characteristic learning set;

based on the feature learning set, a stacked self-encoder is adopted to carry out depth feature coding and feature reconstruction, and a deep relation mining set is generated;

based on the deep relation mining set, performing image and text feature extraction by adopting a convolutional neural network, and performing feature fusion to generate a deep feature set;

based on the deep feature set, a long-time and short-time memory network is adopted to perform time sequence data modeling and relationship prediction, and a non-genetic deep semantic association network is generated;

6. A system for constructing a non-genetic carrier spectrum, characterized in that the system comprises a non-genetic information processing module, a non-genetic space-time model constructing module, a non-genetic multi-modal feature extracting module, a non-genetic knowledge map constructing module and a non-genetic deep learning and modeling module according to the construction method of the non-genetic carrier spectrum of any one of claims 1 to 5.

7. The system for constructing a non-genetic carrier spectrum according to claim 6, wherein the non-genetic information processing module obtains original non-genetic element information including spatiotemporal data by using a crawler technology based on a disclosed target database, and cleans the original non-genetic element information by using a data preprocessing technology to generate a purified non-genetic spatiotemporal data set;

8. The system for constructing a non-genetic carrier spectrum according to claim 6, wherein the non-genetic information processing module comprises a data acquisition sub-module, a data preprocessing sub-module, a geographic position normalization sub-module and a data normalization sub-module;