A personalized recommendation method for remote sensing satellite data

Abstract

The present invention belongs to a recommendation method, and specifically relates to a personalized recommendation method for remote sensing satellite data. It includes the following contents: first, by constructing a satellite resource pool and an expert knowledge base, satellite description information, user information, and prior knowledge between user industries and data types are obtained; then, according to user information and their recent behavior data, a comprehensive recommendation algorithm is used to sort the remote sensing satellite data to obtain a recommended data list; finally, satellite data that meets user needs or is of interest to the user is recommended to the user. The significant effects of the present invention are: (1) Using the remote sensing satellite data recommendation system according to the present invention, the user does not need to provide explicit input. (2) By analyzing the user's historical behavior or field of engagement, data that can meet the user's needs and interests is actively recommended to the user, providing continuous and personalized services, transforming the data interaction mechanism, and solving the real problems of remote sensing satellite data services.

Description

Technical Field

[0001] The invention belongs to a recommendation method, and in particular relates to a personalized recommendation method for remote sensing satellite data. Background Art

[0002] As an important basic resource, remote sensing satellite data is widely used in many fields such as national defense, economy, transportation, energy, and environmental protection. Remote sensing satellite data has the characteristics of massive, multi-source, and heterogeneous. The specific description is as follows: First, satellite data covers a wide range and a long time span. A large amount of historical remote sensing satellite data has been accumulated. At the same time, there are thousands of satellites flying over the earth, carrying various mode payloads, and new remote sensing satellite data is generated every moment. More and more satellites are launched, causing the data volume to explode. Second, remote sensing satellite data comes from a variety of sources, including visible light, infrared, microwave, hyperspectral and other sensor types, and different sensor types, different resolutions, and different band ranges are suitable for different application requirements. Third, satellite data has observation periodicity and can perform long-term time series observations at different angles in the same area. Remote sensing data from different satellite sources differ in many aspects such as storage format, organization method, and metadata standards. Traditional data management methods are difficult to achieve coordinated management of multi-source satellite data and meet personalized data customization needs.

[0003] With the gradual increase in the number of satellites in orbit and the quality of data in my country, the types and quantity of remote sensing satellite data products are increasing, the demand for satellite data is increasing, and the application fields of data are also expanding, which brings great challenges to the storage management and service methods of remote sensing satellite data. The existing remote sensing satellite data storage information is simple, mainly based on basic attributes, and rarely considers and reflects the association between heterogeneous data and the high-level characteristics of the data itself, which is difficult to meet the application needs of high effectiveness. At present, users obtain remote sensing satellite data mainly by relying on simple "metadata" and "artificial experience" to retrieve and discover the required satellite data. This traditional remote sensing satellite data service requires users to have certain professional knowledge, which limits the scope of remote sensing data sharing to a certain extent; and as the volume of data increases, this passive retrieval method is difficult to guarantee the accuracy and timeliness of data, which leads to an increasing imbalance between remote sensing information processing capabilities and data acquisition capabilities. In addition, a large amount of accumulated historical data cannot be effectively utilized, and it has occupied limited storage space for a long time, exacerbating the "overload" of remote sensing satellite data.

[0004] How to obtain the required data quickly and efficiently to meet the needs of users in multiple fields under the condition of data "overload", and how to make use of the historical data accumulated by remote sensing satellites for many years when necessary, are challenges that the remote sensing satellite data storage and management service platform will inevitably face. Summary of the invention

[0005] In view of the defects of the prior art, the present invention provides a personalized recommendation method for remote sensing satellite data, establishes a mechanism for actively discovering effective remote sensing satellite data, and meets the needs of different users more flexibly and efficiently.

[0006] The technical solution of the present invention is: a personalized recommendation method for remote sensing satellite data, which includes the following contents: first, by building a satellite resource pool and an expert knowledge base, satellite description information, user information and prior knowledge between user industries and data types are obtained; then, according to the user information and their recent behavior data, a comprehensive recommendation algorithm is used to sort the remote sensing satellite data to obtain a recommended data list; finally, satellite data that meets the user's needs or is of interest to the user is recommended to the user.

[0007] The above-mentioned remote sensing satellite data personalized recommendation method includes the following steps:

[0008] Step 1: Build a satellite resource pool

[0009] Build a basic database based on satellite attributes, including satellite name, sensor name, sensor type, satellite revisit period, resolution, spectral range, and width;

[0010] Step 2: Build a prior knowledge base

[0011] Construct user-data association information based on prior expert knowledge;

[0012] Step 3: Get user information

[0013] For users, their behavior data within a certain period is obtained, including the area of ​​interest, downloaded data information, browsed data information, and data types of interest, to form a data set;

[0014] Step 4: Build a remote sensing satellite database

[0015] Building a database for satellite data to obtain descriptive information of remote sensing data, file attributes and content attributes of the data, data information, label information and evaluation information, including metadata information extraction and semantic information extraction of remote sensing data, wherein the metadata information extraction is used to obtain the attributes of the data file, including data source, data resolution, data spectrum, imaging time, sensor type, and spatial geographic range; the semantic information extraction is used to obtain the content attributes of the data, including the type of land objects and scene information contained in the data;

[0016] Step 5: Generate recommendation list

[0017] A comprehensive recommendation algorithm is used to prioritize satellite data to obtain a recommendation list.

[0018] Step 6: Display recommendation results

[0019] Display the satellite data information in the recommended list to the user, including basic properties of satellite data, thumbnails and recommended explanations.

[0020] Step 7: Get user feedback

[0021] Get user feedback and make iterative updates.

[0022] A personalized recommendation method for remote sensing satellite data as described above, wherein the certain period in step 3 refers to one month, the range of the area of ​​interest is determined by the longitude and latitude range, and the Path / Row in the WRS-2 global reference system is used for specific statistics, and the longitude and latitude range is converted thereto.

[0023] In the above-mentioned personalized recommendation method for remote sensing satellite data, the downloaded data information and the browsed data information in step 3 both refer to satellite data. Specifically, if the data has been downloaded once, it is determined that the user is interested in the data, and if the data has been browsed three times or more, it is determined that the user is interested in the data.

[0024] In the above-mentioned personalized recommendation method for remote sensing satellite data, step 5 includes the following contents:

[0025] Step 5.1, content-based recommendation algorithm

[0026] The content-based recommendation algorithm extracts the content description features of satellite data according to the similarity of data content attributes to realize similarity calculation. Based on the content-based recommendation algorithm, the data content is characterized by the file attributes and content attributes obtained in step 4. The TF-IDF algorithm is used to express the characteristics of the data content by the data label. The similarity is calculated by the cosine similarity formula of the feature vector:

[0027]

[0028] Among them, w ij is the similarity between data, d i The feature vector representing data i, d j The feature vector representing data j is matched with the personal information of user u and the data features to measure the interest of a specific user in satellite data. The user personal information is obtained from the user information obtained in step 3. The recommended data list L for user u is obtained by sorting them in order of interest. content ,

[0029] Step 5.2, user-based collaborative filtering algorithm

[0030] The collaborative filtering algorithm based on users recommends data that other users with similar interests like to the user, and measures the similarity of interests between users u and v through the Jaccard formula.

[0031]

[0032] Among them, w uv is the interest similarity between users. D(u) indicates the data set that user u is interested in, and D(v) indicates the data set that user v is interested in. D(u) and D(v) are obtained by statistics of user u and user v through step 3 respectively. The data of interest of the K users most similar to the user are recommended. The measurement formula is as follows:

[0033]

[0034] M(u,K) represents K users similar to the user, M(i) represents the set of users interested in data i, q vi Represents user v's interest in data i, and sorts them in order according to their interest to obtain the recommended data list L for user u based on the user-based collaborative filtering algorithm user ,

[0035] Step 5.3, item-based collaborative filtering algorithm

[0036] The item-based collaborative filtering algorithm recommends data similar to the data that the user was interested in before, and measures the similarity between the data through user behavior.

[0037]

[0038] Where |M(i)| represents the number of users interested in data i, and |M(j)| represents the number of users interested in data j.

[0039] User u's interest measurement formula for data i

[0040]

[0041] Among them, D(j,K) is the K data sets that are most similar to the data, D(u) is the data set that the user is interested in, and w ij is the similarity between data, q uj Indicates user u's interest in data j, and sorts them in order of interest to obtain the recommended data list L for user u based on the item-based collaborative filtering algorithm content ,

[0042] Step 5.4, TOP-N recommended data list

[0043] The recommendation data of multiple recommendation algorithms are merged and sorted according to weights and priorities to form the final recommendation results. First, the recommendation data list L obtained by the three recommendation algorithms for a specific user u is content , L user and L content The data interest levels are normalized, and then the data lists are merged and sorted by algorithm weighting and data priority, and finally the TOP-N recommended data list L is obtained. The priority of each type of satellite data is obtained by the association between user industry and data in the prior knowledge base in step 2.

[0044] The significant effects of the present invention are: (1) the remote sensing satellite data recommendation system according to the present invention does not require explicit input from the user. (2) by analyzing the user's historical behavior or field of work, the system actively recommends data that can meet the user's needs and interests, provides continuous and personalized services, transforms the data interaction mechanism, and solves the real problems of remote sensing satellite data services. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention is a flowchart of a personalized recommendation method for remote sensing satellite data. DETAILED DESCRIPTION

[0046] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described with reference to the accompanying drawings.

[0047] In this embodiment, Figure 1 As shown, a remote sensing satellite data personalized recommendation method, the method mainly includes the following steps:

[0048] Step 1: Build a satellite resource pool: Build a basic database based on satellite attributes, including satellite name, sensor name, sensor type, satellite revisit period, resolution, spectral range, width, etc.

[0049] The data of the above seven dimensions are derived from satellite attributes. When the targeted satellite changes, the data of the above seven dimensions need to be updated. Otherwise, once the satellite is determined, the data of the above seven dimensions will no longer change. Of course, there are many dimensions of satellite attributes, such as orbital altitude, inclination, imaging mode, etc., but these dimensions have little to do with the subsequent calculations of this application, and are therefore not listed in this step. If a technician in this field needs to add data of relevant dimensions and make them participate in subsequent calculations, then relevant data can be easily extracted from the attribute data of the satellite.

[0050] Step 2: Build a prior knowledge base: Build user-data association information based on prior expert knowledge, and associate industry users with satellite data. The satellite data requirements of different industry applications have different emphases. The construction of the expert prior knowledge base is mainly used to alleviate the problem of missing user ratings for remote sensing satellite data.

[0051] The associated information in this step is not scored or calculated, but is associated with the data type and the user industry based on expert knowledge. For example, for users studying pest monitoring in the agricultural field, near-infrared satellite data is more in line with user needs; while users studying crop yield estimation prefer multispectral or panchromatic satellite data. Professional technicians in related fields can easily establish the association between data types and their fields.

[0052] Prior knowledge can be used to filter and prioritize the satellite data in step 5.

[0053] Step 3, obtain user information: For a specific user, obtain their behavior data within a certain period, including the area of ​​interest, downloaded data information, browsed data information, data types of interest, etc. The user's preferences for area, time series, and data type are combined with their registration information (occupation, research field, etc.).

[0054] The certain period is one month, and the area of ​​interest is determined by the latitude and longitude range. When performing specific statistics, the Path / Row in the WRS-2 global reference system is used to convert the latitude and longitude range thereto.

[0055] "Downloaded data information and browsed data information" all refer to satellite data. If a user has downloaded or browsed the satellite data, it means that the specific user is interested in the data, and it also means that the users who are interested in the data include the user. The specific quantitative judgment scale is: if it has been downloaded once, it means that the user is interested in the data, and if it has been browsed 3 times or more, it means that the user is interested in the data.

[0056] Statistics are collected for all data of interest to a specific user in a certain period of time. The statistical dimensions include satellite name, sensor name, data resolution, spatial range, and data acquisition time. If the satellite data of interest is relatively concentrated, for example, among the 100 satellite data of interest, the number of data in the spatial range set in Path / Row (123 / 32) is 90 (greater than >70%), it means that the user is interested in a specific area.

[0057] Registration information usually includes user name, nickname, gender, contact information, address, occupation, research field and others. Registration information can be imported into the database through external information or obtained through user self-registration through the interactive interface.

[0058] Step 4: For massive satellite data, build a remote sensing satellite database: obtain the description information of remote sensing data, the file attributes and content attributes of the data, data information, label information and evaluation information. This includes the extraction of metadata information and semantic information of remote sensing data. The metadata information extraction is used to obtain the attributes of the data file, including data source, data resolution, data spectrum, imaging time, sensor type, spatial geographic range, etc.; the semantic information extraction is used to obtain the content attributes of the data, including the type of objects contained in the data, scene information, etc.

[0059] In a specific embodiment of the present invention, the file attributes of the data are generated by cataloging the satellite level 0 data, and the file attribute information is extracted from the metadata file using XML parsing technology, including the following information: sensor, data acquisition time, center point longitude and latitude, four corner longitude and latitude, center sun altitude angle and azimuth, cloud cover, etc.

[0060] In a specific embodiment of the present invention, the content attributes of the data are generated by classifying the satellite product data, and the high-level semantic information of the satellite product data is extracted using a deep convolutional neural network method to obtain the ground object categories in the satellite data, including the following ground objects and scene categories: farmland, forest, grassland, river, lake, residential area, airport, airplane, highway, etc.

[0061] Step 5, generating a recommendation list: using a comprehensive recommendation algorithm to prioritize satellite data to obtain a recommendation list. The comprehensive data recommendation algorithm includes a content-based recommendation algorithm, a user-based collaborative filtering algorithm, and an item-based collaborative filtering algorithm.

[0062] Step 5.1, content-based recommendation algorithm

[0063] The content-based recommendation algorithm extracts the content description features of satellite data according to the similarity of data content attributes to achieve similarity calculation. Based on the content-based recommendation algorithm, the data content is characterized by the file attributes and content attributes obtained in step 4. The TF-IDF algorithm is used to express the characteristics of the data content by the data tags, and the similarity is calculated by the cosine similarity formula of the feature vector:

[0064]

[0065] Among them, w ij is the similarity between data, d i The feature vector representing data i, d jRepresents the feature vector of data j. Match the personal information of user u with the data features to measure the interest of a specific user in satellite data. The user's personal information is obtained from the user information obtained in step 3. At the same time, in order to improve the recommendation efficiency, the data range can be narrowed down with the help of the prior knowledge in step 2. The recommended data list L for user u by the content-based recommendation algorithm is obtained by sorting them in order of interest. content .

[0066] Step 5.2, user-based collaborative filtering algorithm

[0067] The collaborative filtering algorithm based on users recommends data that other users with similar interests like to the user. The Jaccard formula can be used to measure the similarity of the interests of users u and v.

[0068]

[0069] Among them, w uv is the interest similarity between users, D(u) indicates the data set that user u is interested in, and D(v) indicates the data set that user v is interested in. D(u) and D(v) are obtained by statistics of user u and user v through step 3 respectively. Recommend the data that the K users who are most similar to the user are interested in. The measurement formula is as follows:

[0070]

[0071] M(u,K) represents K users similar to the user, M(i) represents the set of users interested in data i, q vi Represents user v's interest in data i. Sort by interest to get the recommended data list L for user u based on the user-based collaborative filtering algorithm user .

[0072] Step 5.3, item-based collaborative filtering algorithm

[0073] The item-based collaborative filtering algorithm recommends data similar to the data that the user was interested in before, and measures the similarity between the data through user behavior.

[0074]

[0075] Where |M(i)| represents the number of users interested in data i, and |M(j)| represents the number of users interested in data j.

[0076] User u's interest measurement formula for data i

[0077]

[0078] Among them, D(j,K) is the K data sets that are most similar to the data, D(u) is the data set that the user is interested in, and w ij is the similarity between data, q uj Represents user u's interest in data j. Sort by interest to get the recommended data list L for user u based on the collaborative filtering algorithm based on items. content .

[0079] Step 5.4, TOP-N recommended data list

[0080] The recommendation data of multiple recommendation algorithms are merged and sorted according to weight and priority to form the final recommendation result. First, the recommendation data list L obtained by the three recommendation algorithms for a specific user u is content , L user and L content The data interest level is normalized, and then the data list is merged and sorted by algorithm weighting and data priority, and finally the TOP-N recommended data list L is obtained. The priority of each type of satellite data is obtained by the association between user industry and data in the prior knowledge base in step 2.

[0081] Step 6, display of recommendation results: display the satellite data information in the recommendation list to the user, including basic attributes of the satellite data, thumbnails and recommendation explanations.

[0082] This step may also add a recommendation explanation, which refers to the reason why the data is recommended, specifically, the recommended features of the recommended data. The recommended features refer to the characteristics of the satellite data that led to it being included in the recommendation list in step 5 to meet user needs, such as high resolution, near-infrared data, longitude and latitude range, etc.

[0083] Step 7, obtaining user feedback, mainly obtains the interaction between the user and the recommendation results, such as whether the user browses or downloads the recommended data. The recommendation results are updated and optimized based on user feedback, and the recommendation list removes the data that the user has downloaded or browsed, and is updated every 5 minutes.

Claims

1. A personalized recommendation method for remote sensing satellite data, characterized by: It includes the following contents: first, by building a satellite resource pool and expert knowledge base, satellite description information, user information, and prior knowledge between user industries and data types are obtained; Then, based on the user information and their recent behavior data, a comprehensive recommendation algorithm is used to sort the remote sensing satellite data to obtain a recommended data list; finally, the satellite data that meets the user's needs or the user's interest is recommended to the user; The steps include: Step 1: Build a satellite resource pool Build a basic database based on satellite attributes, including satellite name, sensor name, sensor type, satellite revisit period, resolution, spectral range, and width; Step 2: Build a prior knowledge base Construct user-data association information based on prior expert knowledge; Step 3: Get user information For users, their behavior data within a certain period is obtained, including the area of ​​interest, downloaded data information, browsed data information, and data types of interest, to form a data set; Step 4: Build a remote sensing satellite database Building a database for satellite data to obtain descriptive information of remote sensing data, file attributes and content attributes of the data, data information, label information and evaluation information, including metadata information extraction and semantic information extraction of remote sensing data, wherein the metadata information extraction is used to obtain the attributes of the data file, including data source, data resolution, data spectrum, imaging time, sensor type, and spatial geographic range; the semantic information extraction is used to obtain the content attributes of the data, including the type of land objects and scene information contained in the data; Step 5: Generate recommendation list A comprehensive recommendation algorithm is used to prioritize satellite data to obtain a recommendation list. Step 6: Display recommendation results Display the satellite data information in the recommended list to the user, including basic properties of satellite data, thumbnails and recommended explanations. Step 7: Get user feedback Get user feedback and make iterative updates; The step 5 includes the following contents: Step 5.1, content-based recommendation algorithm The content-based recommendation algorithm extracts the content description features of satellite data according to the similarity of data content attributes to realize similarity calculation. Based on the content-based recommendation algorithm, the data content is characterized by the file attributes and content attributes obtained in step 4. The TF-IDF algorithm is used to express the characteristics of the data content by the data label. The similarity is calculated by the cosine similarity formula of the feature vector: in, is the similarity between data, Representation data The characteristic vector of Representation data The feature vector of the user The user's personal information is matched with the data features to measure the interest of a specific user in satellite data. The user's personal information is obtained from the user information obtained in step 3. The content-based recommendation algorithm is used to sort the user information according to the interest level. Recommended data list , Step 5.2, user-based collaborative filtering algorithm The collaborative filtering algorithm based on users recommends data that other users like and have similar interests to users, and uses the Jaccard formula to measure users and Similarity of interests in, is the interest similarity between users, Identify the user The dataset of interest is Indicates user The dataset of interest is and For users and users According to the statistics in step 3, the most similar recommendations are The data that a user is interested in is measured by the following formula: Represents users who are similar Users, Indicates data Interested user set, Representative User Data The interests of users are sorted in order according to their interest levels to obtain the user-based collaborative filtering algorithm. Recommended data list , Step 5.3, item-based collaborative filtering algorithm The item-based collaborative filtering algorithm recommends data similar to the data that the user was interested in before, and measures the similarity between the data through user behavior. ， in Represents data Number of interested users, Represents data Number of interested users, user Data The interest measure formula in, is most similar to the data Datasets, is the dataset that the user is interested in, is the similarity between data, Indicates user Data The interests of users are sorted in order according to their interest levels to obtain the collaborative filtering algorithm based on items. Recommended data list , Step 5.4, TOP-N recommended data list The recommendation data of multiple recommendation algorithms are merged and sorted according to weights and priorities to form the final recommendation results. First, the three recommendation algorithms are used for specific users. The recommended data list obtained , and The data interest level is normalized, and then the data list is merged and sorted by algorithm weighting and data priority, and finally the TOP-N recommended data list is obtained. ,The priority of various types of satellite data is obtained through the association between the user industry and the data in the prior knowledge base in step 2.

2. A remote sensing satellite data personalized recommendation method as claimed in claim 1, characterized in that: The certain period in step 3 is one month, and the area of ​​interest is determined by the latitude and longitude range. When performing specific statistics, the Path / Row in the WRS-2 global reference system is used to convert the latitude and longitude range.

3. A remote sensing satellite data personalized recommendation method as claimed in claim 2, characterized in that: The downloaded data information and browsed data information in step 3 all refer to satellite data. If the data has been downloaded once, it is determined that the user is interested in the data. If the data has been browsed three times or more, it is determined that the user is interested in the data.

Images