Remote sensing process recommendation method based on end-to-end structure and internal attribute features

Abstract

The invention relates to the technical field of data processing, in particular to a remote sensing process recommendation method based on an end-to-end structure and internal attribute features, which comprises the following steps: acquiring a remote sensing process to be recommended to obtain a tensor set and a feature attribute set; performing similarity matching on the tensor set of the remote sensing process and all tensor sets in the process library to generate a similarity matrix based on an end-to-end structure; performing similarity calculation on the feature attribute set of the remote sensing process and feature attribute sets of other processes in the process library to generate a similarity matrix based on internal parameters; according to the similarity matrix based on the end-to-end structure and the similarity matrix based on the internal parameters, a preset number of processes with the highest similarity with the remote sensing process are generated, and then the selected processes are recommended to the user. According to the method, data mining can be further carried out on the existing remote sensing algorithm process library, the data capacity is expanded, and the remote sensing algorithm process is quickly and accurately constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a remote sensing process recommendation method based on end-to-end structure and internal attribute features. Background Art

[0002] As a basic spatial data carrier, remote sensing image data can provide a variety of information in a timely and reliable manner, and plays an irreplaceable role in many application fields such as surveying and mapping, land, agriculture, forestry, environment, transportation, meteorology and ocean. These applications are typical long-chain applications, from the original image to the final thematic product that can provide knowledge value, it must go through multiple processing links to form an application process.

[0003] Remote sensing workflow components include algorithm components and process components, which should have complete arrow links and their own basic properties, and the data is mainly stored in a relational database. Among them, an algorithm component is the encapsulation and abstraction of a remote sensing algorithm, which should contain several inputs, outputs and a processing unit; a process component is the encapsulation and abstraction of an application process, which describes the execution logic of multiple algorithms, and its internal components are the encapsulation and abstraction of multiple algorithm components.

[0004] Image data can be used to produce remote sensing products through process flows. As remote sensing products become increasingly complex, customization of remote sensing processes using non-standard algorithms plays an important role in product promotion and production. However, the traditional customized workflow model requires developers and users to have extensive business and technical knowledge, and the configuration process is often complex, resulting in inefficient construction of remote sensing workflows and difficulty in effective promotion. Summary of the invention

[0005] In order to solve the technical problem that the configuration process of the above-mentioned existing remote sensing process recommendation is relatively complicated, resulting in low efficiency in building remote sensing workflows, the purpose of the present invention is to provide a remote sensing process recommendation method based on end-to-end structure and internal attribute features. The technical solution adopted is as follows:

[0006] An embodiment of the present invention provides a remote sensing process recommendation method based on end-to-end structure and internal attribute features, the method comprising the following steps:

[0007] Obtain remote sensing processes to be recommended;

[0008] Using the remote sensing process to generate a tensor set containing an end-to-end structure, and collecting internal parameter information and characteristic attributes of the remote sensing process to construct a characteristic attribute set of the remote sensing process;

[0009] Convert all processes in the process library into tensor sets, make similarity matching between the tensor set of the remote sensing process and all tensor sets in the process library, and generate a similarity matrix based on the end-to-end structure;

[0010] Calculating similarity between the characteristic attribute set of the remote sensing process and the characteristic attribute sets of other processes in the process library to generate a similarity matrix based on internal parameters;

[0011] According to the similarity matrix based on the end-to-end structure and the similarity matrix based on the internal parameters, a preset number of processes with the highest similarity to the remote sensing process are generated, and then the selected processes are recommended to the user.

[0012] Furthermore, the obtaining of the remote sensing process to be recommended includes:

[0013] Configure or upload the newly created algorithm unit or process component, fill in basic parameter information, create incomplete input and output, connect the algorithm unit or process component and the input and output with arrows, and generate a basic process model as a remote sensing process.

[0014] Furthermore, the construction process of the basic process model includes:

[0015] In the remote sensing algorithm process building software, build algorithm units or process components by dragging cloud components with the mouse or adding input and output to processing nodes;

[0016] Based on the algorithm units or process components, the output of the previous process is linked to the input of the next process through different connection methods, generating a basic process model guided by arrows.

[0017] Further, the generating a set of tensors including an end-to-end structure by using the remote sensing process includes:

[0018] Obtain a remote sensing process structure diagram; for each node in the remote sensing process structure diagram, determine whether the current node is a processing node; determine the tensor expression of each processing node according to the determination result of the current node, store all tensor expressions in a tensor set, and obtain an initial tensor set;

[0019] For the initial tensor set, a tensor expression whose input set is a file type is selected as a target tensor expression; the corresponding real tensor is searched according to the realTensorUuid in the target tensor expression, and if the search is successful, the target tensor expression and the real tensor are merged to obtain a new tensor expression;

[0020] The set of several new tensor expressions is taken as a tensor set containing an end-to-end structure.

[0021] Furthermore, the tensor expression of each processing node is determined according to the judgment result of the current node, including:

[0022] When the current node is a processing node, determine whether the processing node has a predecessor node and a successor node. If the processing node has a predecessor node or a successor node, store the existing node into the tensor expression;

[0023] Traverse the next node until all nodes are traversed, and determine the tensor expression of each processing node.

[0024] Furthermore, the definition of the initial tensor set includes:

[0025] First definition: Let a tensor expression be denoted as F, where F = {I, O, L}, where I represents the input tensor sequence, O represents the output tensor sequence, and L represents the nesting level;

[0026] Second definition: Based on the first definition, for the remote sensing process algorithm, a mapping relationship from input to output is constructed: O→G(I; L) represents the mapping of output sequence to input sequence, and I→G(O; L) represents the mapping of input sequence to output sequence;

[0027] Third definition: Define the initial tensor set:

[0028] Among them, I i Represents the input tensor sequence corresponding to processing node i. The input tensor sequence contains multiple input tensors. n represents the number of input tensors; O i Represents the output tensor sequence corresponding to processing node i. The output tensor sequence contains multiple output tensors. m represents the number of output tensors, Z + Represents a positive integer.

[0029] Furthermore, the process of performing similarity matching between the tensor set of the remote sensing process and all tensor sets in the process library to generate a similarity matrix based on an end-to-end structure includes:

[0030] Using cosine similarity, the similarity between the tensor set of the remote sensing process and each tensor set in the process library is calculated in turn, and all similarities are constructed into a similarity matrix of an end-to-end structure.

[0031] Furthermore, the similarity calculation is performed on the characteristic attribute set of the remote sensing process and the characteristic attribute sets of other processes in the process library to generate a similarity matrix based on internal parameters, including:

[0032] The process in the process library with the same classification attribute as the remote sensing process is taken as the matching process, and the process internal attribute set and the tensor basic attribute set in the feature attribute set of the remote sensing process and each matching process are obtained;

[0033] Similarity analysis is performed based on the process internal attribute set in the feature attribute set of the remote sensing process and each matching process to obtain the basic attribute feature similarity matrix;

[0034] Perform similarity matching on the tensor basic attribute sets in the remote sensing process and the feature attribute sets of each matching process to obtain a tensor parameter feature similarity matrix;

[0035] The basic attribute feature similarity matrix and the tensor parameter-based feature similarity matrix are combined to obtain an internal parameter-based similarity matrix.

[0036] Furthermore, similarity analysis is performed on the process internal attribute set in the feature attribute set of the remote sensing process and each matching process to obtain a basic attribute feature similarity matrix, including:

[0037] For the string type data in the internal attribute set of the process, feature engineering methods are used to extract features of the string type data, and then feature matching of the corresponding attributes is performed on the extracted features to obtain matching feature pairs of the remote sensing process and matching feature pairs of each matching process;

[0038] Similarity matching is used to calculate the similarity of the matching feature pairs of the remote sensing process and the matching feature pairs of each matching process, and then the final weighted summation is performed to obtain the basic attribute feature similarity matrix.

[0039] Furthermore, generating a preset number of processes with the highest similarity to the remote sensing process according to the similarity matrix based on the end-to-end structure and the similarity matrix based on the internal parameters includes:

[0040] The similarity matrix based on the end-to-end structure is multiplied by the similarity matrix based on the internal parameters to obtain the similarity value between the remote sensing process and other processes in the process library;

[0041] The similarity values ​​are arranged in descending order to obtain the processes in the process library corresponding to the first preset number of similarity values ​​in the arranged sequence.

[0042] The present invention has the following beneficial effects:

[0043] The present invention provides a remote sensing process recommendation method based on end-to-end structure and internal attribute features. The method obtains the remote sensing process to determine the tensor set and feature attribute set of the end-to-end structure, matches the tensor set of the remote sensing process with all tensor sets in the process library, generates a similarity matrix based on the end-to-end structure, calculates the similarity between the feature attribute set of the remote sensing process and the feature attribute set of other processes in the process library, generates a similarity matrix based on internal parameters, and then generates a preset number of processes with the highest similarity to the remote sensing process, and recommends them to users. The present invention can deeply explore the input-output relationship of process components, which helps to further explore potential data relationships in the existing process library and expand data capacity; secondly, the present invention can help users design high-quality remote sensing workflow models, improve the construction efficiency of remote sensing workflows to a certain extent, and then improve the reliability of similarity calculations between remote sensing algorithm processes, find similar workflows, so as to recommend the best matching application driver nodes to users. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A first flow chart of a remote sensing process recommendation method based on end-to-end structure and internal attribute features according to an embodiment of the present invention;

[0046] Figure 2 A second flow chart of a remote sensing process recommendation method based on end-to-end structure and internal attribute features according to an embodiment of the present invention;

[0047] Figure 3 A structural diagram of an algorithm in a process library in a remote sensing process recommendation method based on end-to-end structure and internal attribute features according to an embodiment of the present invention;

[0048] Figure 4 This is a structural diagram of a process in a process library in a remote sensing process recommendation method based on end-to-end structure and internal attribute features in one embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0050] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0051] This embodiment provides a remote sensing process recommendation method based on end-to-end structure and internal attribute features. The first flow chart of the remote sensing process recommendation method is as follows: Figure 1 As shown, the second flowchart of the remote sensing process recommendation method is as follows Figure 2 As shown, the following steps are included:

[0052] S1, obtaining the remote sensing process to be recommended.

[0053] In this embodiment, a newly created algorithm unit or process component is configured or uploaded, basic parameter information is filled in, incomplete input and output are created, and the algorithm unit or process component and the input and output are connected with arrows to generate a basic process model as a remote sensing process. Incomplete input and output refers to no input, incomplete input, no output, or incomplete output.

[0054] As a specific implementation method, the detailed construction steps of the above basic process model include:

[0055] The first step is to build the algorithm unit or process component in the remote sensing algorithm process building software by dragging the cloud components with the mouse or adding input and output to the processing nodes.

[0056] The second step is to link the output of the previous process to the input of the next process through different connection methods based on the algorithm unit or process component, and generate a basic process model guided by arrows.

[0057] Specifically, the remote sensing algorithm process building software can be the National Remote Sensing Data and Application Service Platform, in which algorithm modules or process components are built by dragging cloud components with the mouse or adding inputs and outputs to processing nodes; for the process, the output of the previous process is linked to the input of the next process through sequential, parallel and other logical methods, and finally a process model guided by arrows and similar to a neural network structure is generated as the basic process model.

[0058] So far, this embodiment has obtained the remote sensing process to be recommended.

[0059] S2, uses the remote sensing process to generate a tensor set containing an end-to-end structure, collects the internal parameter information and characteristic attributes of the remote sensing process, and constructs a characteristic attribute set of the remote sensing process.

[0060] In this embodiment, the characteristic attribute set is recorded as P, P = (L f ,B att ,T att ,L other ,L), where L f represents the set of expressions generated according to the end-to-end structure, B att , represents the basic attribute set contained in the process algorithm or process, T att Indicates L f The attribute set of the corresponding tensor, L other represents a set of other descriptive attributes, and L represents the execution level.

[0061] Definition 1, Definition 2, and Definition 3, i.e., tensor set definitions, are specific implementation methods. The steps for determining a tensor set of an end-to-end structure include:

[0062] The first step is to obtain the remote sensing process structure diagram; for each node in the remote sensing process structure diagram, determine whether the current node is a processing node; according to the judgment result of the current node, determine the tensor expression of each processing node, store all tensor expressions in the tensor set, and obtain the initial tensor set.

[0063] Specifically, the essence of the remote sensing process structure diagram is a directed acyclic graph. For each node in the directed acyclic graph, first perform a breadth search traversal from the node without a predecessor node, but before that, observe whether the current node is a processing node. When the current node being traversed is a processing node, observe the predecessor and successor nodes of the processing node; when the processing node has a predecessor node (input) or a successor node (output), store the predecessor and successor nodes in the tensor expression according to the rules defined by the tensor set, and set the execution level L to 1; then traverse the next node until all nodes in the directed acyclic graph have been traversed and analyzed, at which time each processing node has its corresponding tensor expression.

[0064] It is worth mentioning that for an algorithm component, it has only one attribute set, several input nodes, one processing node and several output nodes. Accordingly, an algorithm component has only one tensor expression; while for a process component, it contains several algorithm components, that is, there are several processing nodes, each processing node contains several inputs and outputs, and accordingly, a process component has multiple tensor expressions, that is, there is a tensor set, which is used to store a set of several tensor expressions as the initial tensor set.

[0065] The definition of the initial tensor set includes:

[0066] Definition 1: Let a tensor expression be denoted as F, F = {I, O, L}, where I represents the input tensor sequence, O represents the output tensor sequence, and L represents the nesting level, which specifically represents the number of execution layers of the remote sensing algorithm.

[0067] Definition 2: Based on Definition 1, for the remote sensing process algorithm, a mapping relationship from input to output is constructed: O→G(I; L) represents the mapping of output sequence to input sequence, and I→G(O; L) represents the mapping of input sequence to output sequence.

[0068] Definition 3: Define the initial tensor set:

[0069] Among them, I i Represents the input tensor sequence corresponding to processing node i. The input tensor sequence contains multiple input tensors. n represents the number of input tensors; O i Represents the output tensor sequence corresponding to processing node i. The output tensor sequence contains multiple output tensors. m represents the number of output tensors, Z + Represents a positive integer.

[0070] In the second step, for the initial tensor set, select the tensor expression whose input set is the file type as the target tensor expression; search for the corresponding real tensor according to the realTensorUuid in the target tensor expression. If the search is successful, merge the target tensor expression and the real tensor to obtain a new tensor expression.

[0071] In this embodiment, after obtaining the initial tensor set, each tensor expression in the initial tensor set is traversed. For the input set of the remote sensing process, the input part in the expression is divided into two categories: file type and numerical type, and the output part only has the file type; if there is a file type in the input, the corresponding real tensor is searched according to the realTensorUuid in the basic properties of the tensor. If the search is successful, the target tensor expression and the real tensor are merged, and the nesting level L is set to 1.

[0072] Among them, realTensorUuid represents the real ID of the tensor. The remote sensing processing process is similar to the common flowchart, which is a chain structure, but there are too many processes, which leads to excessive vertical extension. In order to flatten the description between processing nodes horizontally and make the visual arrangement more intuitive, this attribute is added to the description of the tensor. Specific example: an output tensor of processing node A is A_O1, and an input tensor of processing node B is B_I1. If the output tensor A_O1 of processing node A is the input of the next processing node B, then A_O1 and B_I1 are actually an instance, but in the xml description text, these two inputs and outputs are separated, that is, the realTensorUuid of processing node B = A_O1.

[0073] The third step is to take the set of several new tensor expressions as a tensor set containing an end-to-end structure.

[0074] In this embodiment, the second step of the above step S2 is repeated until each target tensor expression is traversed, and a complete tensor set about the remote sensing process is obtained, that is, a tensor set containing an end-to-end structure is obtained.

[0075] It is worth mentioning that tensor expressions store the logical relationship between input and output in an algorithm or process, while the basic properties and other additional content of the algorithm or process during the construction process are stored in a relational database. The data is stored in plain text to represent the corresponding actual meaning.

[0076] So far, this embodiment has obtained a tensor set containing an end-to-end structure corresponding to the remote sensing process.

[0077] S3, converts all processes in the process library into tensor sets, makes similarity matching between the tensor sets of the remote sensing process and all tensor sets in the process library, and generates a similarity matrix based on the end-to-end structure.

[0078] Specifically, referring to the implementation process of step S2 above, all algorithms and processes in the process library are converted into tensor sets corresponding to the remote sensing process, and tensor sets with different data sizes in the set are obtained; feature engineering and word embedding methods are used to extract features of the input set and output set in the tensor set respectively, and then similarity calculation is performed based on the extracted features, so as to match the tensor set of the remote sensing process with the tensor set of all processes in the process library.

[0079] It should be noted that the input and output parts of the tensor have the following types: no input, incomplete input, no output, and incomplete output; if there is no input, the input part does not participate in the similarity calculation, and if there is no output, the output part does not participate in the similarity calculation.

[0080] Among them, when performing similarity matching analysis, cosine similarity is used to calculate the similarity between the tensor set of the remote sensing process and each tensor set in the process library in turn, and all similarities are constructed into a similarity matrix of an end-to-end structure.

[0081] As an example, the calculation formula for the similarity between the tensor set of the remote sensing process and any tensor set in the process library can be:

[0082] In the formula, Sim(f) represents the similarity, r1 represents the tensor set of the remote sensing process, r2 represents any tensor set in the process library, V r1 The dot product of the embedding vectors of the tensor set representing the remote sensing process, V r2 represents the dot product of the embedding vectors of any set of tensors in the process library, ||V r1 ||The modulus of the embedding vector of the tensor set representing the remote sensing process, ||V r2 || represents the modulus of the embedding vector of any tensor set in the process library.

[0083] Step S3 is described with examples in conjunction with a specific embodiment:

[0084] The structure diagram of an algorithm in the process library is as follows Figure 3 As shown, for the algorithm, there is only one tensor expression, namely, F = {objImg, bandNNbs, DEMImg, SRStype, productName; RPCimg; 1}, and each element in the tensor expression is the tensor name.

[0085] The structure diagram of a process in the process library is as follows Figure 4 As shown, there are a finite number of tensor expressions for the process:

[0086] F1={GF_DNx,bandNbs,DEMImg,SRStype,produtName;GF_HLH0;1}

[0087] F2={GF_HLH0,eoiMetaNorm,includeRegs,excludeRegs,otRegs,mergType;GF_HLH;1}

[0088] F3={GF_HLH,red_NIR_Nbs,radioType,red_NIR_IDs;TOAP;1}

[0089] F4={TOAP,includeRegs,excludeRegs,otRegs,mergType;GFx;1}

[0090] F5=F1∪F2&set L=2→{GF_DNx,bandNbs,DEMImg,SRStype,produtName,

[0091] eoiMetaNorm,includeRegs,excludeRegs,otRegs,mergType;GF_HLH;2}

[0092] F6=F2∪F3&set L=2→{GF_HLH0,eoiMetaNorm,includeRegs,excludeRegs,otRegs,

[0093] mergType,red_NIR_Nbs,radioType,red_NIR_IDs;TOAP;2}

[0094] F7=F3∪F4&set L=2→{GF_HLH,red_NIR_Nbs,radioType,red_NIR_IDs,

[0095] includeRegs,excludeRegs,otRegs,mergType;GFx;2}

[0096] F8=F1∪F2∪F3&set L=3→{GF_DNx,bandNbs,DEMImg,SRStype,produtName,

[0097] eoiMetaNorm,includeRegs,excludeRegs,otRegs,mergType,red_NIR_Nbs,radioType,red_NIR_IDs;TOAP;3}

[0098] F9=F2∪F3∪F4&set L=3→{GF_HLH0,eoiMetaNorm,includeRegs,excludeRegs,

[0099] otRegs,mergType,red_NIR_Nbs,radioType,red_NIR_IDs,includeRegs,excludeRegs,otRegs,mergType;GFx;3}

[0100] F 10 =F1∪F2∪F3∪F4&set L=4→{GF_DNx,bandNbs,DEMImg,SRStype,

[0101] produtName,eoiMetaNorm,includeRegs,excludeRegs,otRegs,mergType,red_NIR_Nbs,radioType,red_NIR_IDs,includeRegs,excludeRegs,otRegs,mergType; GFx; 4}

[0102] Among the above-mentioned finite tensor expressions, F1 to F4 are tensor expressions of a single algorithm or a single node level (node-tensor-sequence), and there can be multiple tensor representation sequences representing a single algorithm from input to output; F 10 It is the maximum tensor representation set (graph-tensor-sequence) of the entire graph level, and there is only one, which represents the maximum tensor representation set finally obtained through the end-to-end method; F5 to F9 are subsequences (nodes-tensor-sequence) obtained from the flowchart, and there can be multiple subsequences. These subsequences are used to find the tensor representation structure existing in the process through mining.

[0103] Then, after obtaining the tensor expressions and tensor sets, the tensor expressions and tensor sets created by the user are obtained, and a similarity matrix based on the end-to-end structure is generated through similarity calculation.

[0104] So far, this embodiment has obtained a similarity matrix based on an end-to-end structure corresponding to the remote sensing process.

[0105] S4, calculate the similarity between the characteristic attribute set of the remote sensing process and the characteristic attribute sets of other processes in the process library, and generate a similarity matrix based on internal parameters.

[0106] As a specific implementation, the step of obtaining a similarity matrix based on internal parameters may include:

[0107] In the first step, the processes in the process library with the same classification attributes as the remote sensing process are taken as matching processes, and the process internal attribute set and tensor basic attribute set in the feature attribute set of the remote sensing process and each matching process are obtained.

[0108] First, the process feature attributes are stored in the corresponding relational database. The attribute of an algorithm or process corresponds to a record in the database component table; and the tensor attributes are stored in the tensor table and physically associated with the component table through foreign keys.

[0109] Among them, some fields in the basic attributes of algorithms or processes are not differentiated. The two are merged into one table to reduce redundancy. The specific basic attribute settings are as follows:

[0110] C_UUID, algorithm / process ID; FlowName, algorithm / process name; FlowChsName, Chinese name; ShowName, display name; FlowDesc, description; FlowVersion, version number; ProName, product name; ProChsName, product Chinese name; ProClass, product classification; ProLevel, product level; Author, author; AuthorID, author ID; IsShared, whether it is shared.

[0111] The basic properties of tensors are stored in the tensor table, which is physically associated with the component table through foreign keys. In addition to basic properties, tensors also have additional configurations such as default values ​​and optional values, which are still physically associated with the tensor table through foreign keys. The specific basic property settings are as follows:

[0112] tUuid, tensor ID; tensorName, tensor English name; tensorChsName, tensor Chinese name; description, description; tensorType, tensor type; tensorSource, tensor source; defaultValue, default value; optionValue, optional value; needLocked, whether to lock; isHide, whether to hide.

[0113] Secondly, obtain the classification of the remote sensing process and the classification of each algorithm or process in the process library. If the two are the same type of processing components, the processes with the same classification attributes are regarded as matching processes for the next step; otherwise, there is no matching meaning.

[0114] It is worth mentioning that the process internal attribute set and the tensor basic attribute set are the basic data for analyzing the matching situation, which can provide data support for the analysis of subsequent steps.

[0115] In the second step, similarity analysis is performed based on the remote sensing process and the process internal attribute set in the feature attribute set of each matching process to obtain the basic attribute feature similarity matrix.

[0116] As a specific implementation, the step of obtaining the basic attribute feature similarity matrix may include:

[0117] In the first sub-step, for the string type data in the internal attribute set of the process, the feature engineering method is used to extract the features of the string type data, and then the feature matching of the corresponding attributes is performed on the extracted features to obtain the matching feature pairs of the remote sensing process and the matching feature pairs of each matching process.

[0118] In the second sub-step, similarity matching is used to calculate the similarity of the matching feature pairs of the remote sensing process and the matching feature pairs of each matching process, and then a final weighted sum is performed to obtain the basic attribute feature similarity matrix.

[0119] In this embodiment, the process internal attribute set of the remote sensing process is matched with the process internal attribute set of each matching process in the process library for similarity. Since most of the remote sensing algorithm process attribute sets are string type data, the feature engineering method is adopted for the string descriptive text, and the numerical type can be directly compared; then, feature extraction is performed on each field in the attribute feature, and the extracted features are matched with the corresponding attributes to obtain L r1 T1, L r1 T2…with L r2 T1, L r2 T2..., in this embodiment, T1 represents C_UUID, T2 represents FlowName, L r1 T1 represents the first attribute T1 of process r1, L r1 T2 represents the second attribute T2 of process r1, L r2 T1 represents the first attribute T1 of process r2, L r2 T2 represents the second attribute T2 of process r2; similarity matching is used to obtain L r1 T1 and L r2 The similarity of T1 is finally weighted and summed to obtain the similarity matrix based on attribute features. Among them, the means to achieve the similarity between two string type data can be cosine similarity. The implementation process of cosine similarity is a prior art and is not within the scope of protection of the present invention. It will not be elaborated in detail here. The specific content of the first attribute T1 and the second attribute T2 can be determined by the specific basic attributes of the process set above. The order of the specific basic attributes of the process can be set by the implementer according to the specific actual situation.

[0120] Among them, the process attributes involved in the calculation can be: C_UUID, algorithm / process ID; FlowName, algorithm / process name; FlowChsName, Chinese name; ShowName, display name; FlowDesc, description; FlowVersion, version number; ProName, product name; ProChsName, product Chinese name; ProClass, product classification; ProLevel, product level; Author, author; AuthorID, author ID; IsShared, whether it is shared.

[0121] The third step is to perform similarity matching on the tensor basic attribute sets in the feature attribute sets of the remote sensing process and each matching process to obtain a tensor parameter feature similarity matrix.

[0122] In this embodiment, for the same type of input or output, different parameter values ​​may be taken, that is, the properties of the tensor itself are also different. Then, the basic property set of the tensor is obtained, and the tensor properties of the two tensor sets are matched for similarity to obtain a similarity matrix based on tensor parameter features.

[0123] Among them, the tensor attributes involved in the operation can be: tUuid, tensor ID; tensorName, tensor English name; tensorChsName, tensor Chinese name; description, description; tensorType, tensor type; tensorSource, tensor source; defaultValue, default value; optionValue, optional value; needLocked, whether to lock; isHide, whether to hide.

[0124] The fourth step is to combine the basic attribute feature similarity matrix and the tensor parameter feature similarity matrix to obtain the internal parameter-based similarity matrix.

[0125] In this embodiment, the basic attribute feature similarity matrix and the tensor parameter feature similarity matrix are respectively assigned the same weight, i.e., 0.5, and the two matrices are combined based on the weights to obtain a new similarity matrix as the similarity matrix based on the internal parameters.

[0126] It is worth noting that the size of the similarity matrix is ​​n*n, and the elements on the diagonal of the matrix are all 1. So far, this embodiment has obtained the similarity matrix based on the internal parameters corresponding to the remote sensing process.

[0127] S5, generating a preset number of processes with the highest similarity to the remote sensing process according to the similarity matrix based on the end-to-end structure and the similarity matrix based on the internal parameters, and then recommending the selected processes to the user.

[0128] Specifically, the similarity matrix based on the end-to-end structure is multiplied by the similarity matrix based on the internal attributes to obtain a similarity result, and a preset number of recommended processes with the highest similarity to the remote sensing process are generated, where the preset number is an integer greater than 1; return to step S1 and repeat the process until the user constructs the final remote sensing process.

[0129] In this embodiment, after all the remote sensing processes created by the user are compared with the processes in the process library, the loop is stopped. Regarding the remote sensing process, for the exemplary process components in the process library, the high-resolution image NDVI processing V1.0 process, the process contains three components, and the id example of the process is Flow_20240523231124886. If the user creates a process called high-resolution image NDVI processing V1.1, the results of the similarity matching between the two meet the requirements of the preset number of recommended processes with the highest similarity to the remote sensing process. At this time, the process with id Flow_20240523231124886 can be recommended to the user; the remaining processes in the process library are analyzed in this way to obtain multiple process components with similarities greater than or equal to the similarity threshold, and all the process components obtained at this time are configured as a complete process to obtain the final remote sensing process.

[0130] As a specific implementation, the implementation steps of step S5 may include:

[0131] In the first step, the similarity matrix based on the end-to-end structure is multiplied by the similarity matrix based on the internal parameters to obtain the similarity value between the remote sensing process and other processes in the process library.

[0132] In the second step, the similarity values ​​are arranged in descending order to obtain the processes in the process library corresponding to the first preset number of similarity values ​​in the arranged sequence.

[0133] It should be noted that the attribute information and parameter information in the algorithm process are stored in the database separately, including: the tensor information is converted into a sequence and stored in the set L f The basic attribute information is stored in the component basic attribute set B att ;Tensor parameter information is stored in parameter set T att , so that the number and types of components in the process library can be continuously expanded.

[0134] In summary, the present invention obtains the remote sensing process to determine the tensor set and feature attribute set of the end-to-end structure, so that the tensor set of the remote sensing process is matched with all tensor sets in the process library for similarity, and a similarity matrix based on the end-to-end structure is generated. The feature attribute set of the remote sensing process and the feature attribute set of other processes in the process library are similarly calculated, and a similarity matrix based on internal parameters is generated, and then a preset number of processes with the highest similarity to the remote sensing process are generated, and they are recommended to users. The present invention can deeply explore the input-output relationship of the process components, which helps to further explore potential data relationships in the existing process library and expand data capacity; secondly, the present invention can help users design high-quality remote sensing workflow models, improve the construction efficiency of remote sensing workflows to a certain extent, and then improve the reliability of similarity calculations between remote sensing algorithm processes, find similar workflows, so as to recommend the best matching application driver nodes to users.

[0135] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A remote sensing process recommendation method based on end-to-end structure and internal attribute features, characterized in that: The following steps are involved: Obtain remote sensing processes to be recommended; Using the remote sensing process to generate a tensor set containing an end-to-end structure, and collecting internal parameter information and characteristic attributes of the remote sensing process to construct a characteristic attribute set of the remote sensing process; Convert all processes in the process library into tensor sets, make similarity matching between the tensor set of the remote sensing process and all tensor sets in the process library, and generate a similarity matrix based on the end-to-end structure; Calculating similarity between the characteristic attribute set of the remote sensing process and the characteristic attribute sets of other processes in the process library to generate a similarity matrix based on internal parameters; According to the similarity matrix based on the end-to-end structure and the similarity matrix based on the internal parameters, a preset number of processes with the highest similarity to the remote sensing process are generated, and then the selected processes are recommended to the user.

2. A remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 1, characterized in that: The step of obtaining the remote sensing process to be recommended includes: Configure or upload the newly created algorithm unit or process component, fill in basic parameter information, create incomplete input and output, connect the algorithm unit or process component and the input and output with arrows, and generate a basic process model as a remote sensing process.

3. The remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 2, characterized in that: The construction process of the basic process model includes: In the remote sensing algorithm process building software, build algorithm units or process components by dragging cloud components with the mouse or adding input and output to processing nodes; Based on the algorithm units or process components, the output of the previous process is linked to the input of the next process through different connection methods, generating a basic process model guided by arrows.

4. The remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 3 is characterized in that: The step of using the remote sensing process to generate a set of tensors having an end-to-end structure includes: Obtain a remote sensing process structure diagram; for each node in the remote sensing process structure diagram, determine whether the current node is a processing node; determine the tensor expression of each processing node according to the determination result of the current node, store all tensor expressions in a tensor set, and obtain an initial tensor set; For the initial tensor set, a tensor expression whose input set is a file type is selected as a target tensor expression; the corresponding real tensor is searched according to the realTensorUuid in the target tensor expression, and if the search is successful, the target tensor expression and the real tensor are merged to obtain a new tensor expression; The set of several new tensor expressions is taken as a tensor set containing an end-to-end structure.

5. The remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 4, characterized in that: The tensor expression of each processing node is determined according to the judgment result of the current node, including: When the current node is a processing node, determine whether the processing node has a predecessor node and a successor node. If the processing node has a predecessor node or a successor node, store the existing node into the tensor expression; Traverse the next node until all nodes are traversed, and determine the tensor expression of each processing node.

6. A remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 5, characterized in that: The definition of the initial tensor set includes: First definition: Let a tensor expression be denoted as F, where F = {I, O, L}, where I represents the input tensor sequence, O represents the output tensor sequence, and L represents the nesting level; Second definition: Based on the first definition, for the remote sensing process algorithm, a mapping relationship from input to output is constructed: O→G(I; L) represents the mapping of output sequence to input sequence, and I→G(O; L) represents the mapping of input sequence to output sequence; Third definition: Define the initial tensor set: Among them, I i Represents the input tensor sequence corresponding to processing node i. The input tensor sequence contains multiple input tensors. n represents the number of input tensors; O i Represents the output tensor sequence corresponding to processing node i. The output tensor sequence contains multiple output tensors. m represents the number of output tensors, Z + Represents a positive integer.

7. The remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 1, characterized in that: The process of performing similarity matching between the tensor set of the remote sensing process and all the tensor sets in the process library to generate a similarity matrix based on an end-to-end structure includes: Using cosine similarity, the similarity between the tensor set of the remote sensing process and each tensor set in the process library is calculated in turn, and all similarities are constructed into a similarity matrix of an end-to-end structure.

8. The remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 1, characterized in that: The similarity calculation is performed on the characteristic attribute set of the remote sensing process and the characteristic attribute sets of other processes in the process library to generate a similarity matrix based on internal parameters, including: The process in the process library with the same classification attribute as the remote sensing process is taken as the matching process, and the process internal attribute set and the tensor basic attribute set in the feature attribute set of the remote sensing process and each matching process are obtained; Similarity analysis is performed based on the process internal attribute set in the feature attribute set of the remote sensing process and each matching process to obtain the basic attribute feature similarity matrix; Perform similarity matching on the tensor basic attribute sets in the remote sensing process and the feature attribute sets of each matching process to obtain a tensor parameter feature similarity matrix; The basic attribute feature similarity matrix and the tensor parameter-based feature similarity matrix are combined to obtain an internal parameter-based similarity matrix.

9. A remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 8, characterized in that: The similarity analysis is performed on the process internal attribute set in the feature attribute set of the remote sensing process and each matching process to obtain a basic attribute feature similarity matrix, including: For the string type data in the internal attribute set of the process, feature engineering methods are used to extract features of the string type data, and then feature matching of the corresponding attributes is performed on the extracted features to obtain matching feature pairs of the remote sensing process and matching feature pairs of each matching process; Similarity matching is used to calculate the similarity of the matching feature pairs of the remote sensing process and the matching feature pairs of each matching process, and then the final weighted summation is performed to obtain the basic attribute feature similarity matrix.

10. The remote sensing process recommendation method based on end-to-end structure and internal attribute features according to claim 1, characterized in that: The generating, according to the similarity matrix based on the end-to-end structure and the similarity matrix based on the internal parameters, a preset number of processes having the highest similarity to the remote sensing process comprises: The similarity matrix based on the end-to-end structure is multiplied by the similarity matrix based on the internal parameters to obtain the similarity value between the remote sensing process and other processes in the process library; The similarity values ​​are arranged in descending order to obtain the processes in the process library corresponding to the first preset number of similarity values ​​in the arranged sequence.

Images