Directional message generation method, apparatus and device, and medium

Abstract

The invention provides a trend message generation method and device, equipment and a medium, and relates to the technical field of remote sensing image processing and natural language processing, and the method comprises the steps: obtaining a satellite time sequence image; extracting image features from the satellite time sequence image; generating text description statements corresponding to the image features, and reordering the text description statements; and generating a trend message based on the reordered text description statements. According to the method and the device, the satellite time sequence image change detection interpretation result is directly converted into character description output which can be understood by people, the timeliness requirement of trend message generation is met, and technical support is provided for military and civilian fields such as urban planning and region / target monitoring.

Description

Technical Field

[0001] The present disclosure relates to the technical fields of remote sensing image processing and natural language processing, and in particular to a method, device, equipment and medium for generating a trend message. Background Art

[0002] Satellite observation is a large-scale and comprehensive means of earth observation. With the significant improvement of the spatial and temporal resolutions of space-based remote sensing satellites, the application efficiency of satellite observation technology in military and civilian fields such as key area observation and ground object change monitoring has become increasingly prominent. Satellite time series images refer to a collection of multiple observation sequence remote sensing images of the same area obtained within a time range. Movement messages refer to formatted text descriptions of changes in the target of interest in a given area. The generation of movement messages based on imaging remote sensing satellite time series data can characterize the movement change characteristics of regions / targets from a temporal dimension. It is an important way to study the laws of activities such as regional / target change pattern discovery and change trend prediction. The relevant results have important application value in the fields of national defense and people's livelihood. Dynamic monitoring of typical regional changes and automatic generation of movement messages are the basis for realizing intelligent interpretation of key areas / targets. However, the current automatic generation technology of movement messages based on satellite time series images has a low degree of automation and strong subjective dependence. It is impossible to directly convert the satellite time series image change detection and interpretation results into human-understandable text description output, and it is difficult to meet the timeliness requirements of movement message generation. Summary of the invention

[0003] In view of the above problems, the present disclosure provides a method, apparatus and device for generating a trend message.

[0004] According to a first aspect of the present disclosure, a method for generating a trend message is provided, comprising: acquiring satellite time-series images; extracting image features from the satellite time-series images; generating text description sentences corresponding to the image features, and reordering the text description sentences; and generating a trend message based on the reordered text description sentences.

[0005] According to an embodiment of the present disclosure, before extracting image features of satellite time series images from satellite time series images, the method also includes: preprocessing the satellite time series images; wherein the satellite time series images include image names and image header information; extracting the image shooting time of the satellite time series images from the image names; and extracting the latitude and longitude of the image shooting location of the satellite time series images from the image header information.

[0006] According to an embodiment of the present disclosure, extracting image features of satellite time-series images from satellite time-series images includes: obtaining a feature vector corresponding to each satellite time-series image; subtracting two consecutive feature vectors to generate a feature vector difference; and inputting the feature vector difference into a long short-term memory network to obtain encoded image features.

[0007] According to an embodiment of the present disclosure, the encoded image features are obtained by inputting the feature vector difference into a long short-term memory network, including: weighting the image features to obtain a context vector of the satellite time series image; and inputting the context vector into a t-way long short-term memory network in sequence to generate a corresponding text description sentence.

[0008] According to an embodiment of the present disclosure, context vectors of satellite time series images are sequentially input into a t-way long short-term memory network to generate corresponding text description sentences, including: inputting the context vector corresponding to the i-th moment into a word vector model to obtain a corresponding word vector; inputting the context vector corresponding to the i-th moment, the word vector corresponding to the context vector, and the hidden state vector corresponding to the i-1-th moment into a t-way long short-term memory network; and generating a hidden state vector corresponding to the i-th moment.

[0009] According to an embodiment of the present disclosure, after generating a hidden state vector corresponding to the i-th moment, the method includes: using a normalized exponential function to calculate the hidden state vector at the i-th moment to obtain the probability distribution of the predicted word at the i-th moment; using a cross entropy function to calculate the probability distribution at the i-th moment to obtain the loss between the probability distribution and the actual label; using the loss to perform update training; using the word corresponding to the maximum probability index position in the probability distribution as the predicted word at the i-th moment; and concatenating the words at all moments to form a complete sentence.

[0010] According to an embodiment of the present disclosure, generating a trend message based on the reordered text description sentences includes: generating the trend message according to a message-specific format; and outputting the trend message according to a template.

[0011] A second aspect of the present disclosure provides a method and device for generating a trend message, comprising: an acquisition module, used to acquire satellite time-series images; an extraction module, used to extract image features of the satellite time-series images from the satellite time-series images; an image text description sentence module, used to generate a text description sentence corresponding to the image features; a sorting module, used to re-sort the text description sentences; and a trend message generation module, used to generate a trend message based on the re-sorted text description sentences.

[0012] A third aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above method.

[0013] The fourth aspect of the present disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the above method.

[0014] According to a trend message generation method, device and equipment provided by the present invention, it is possible to directly convert the satellite time-series image change detection interpretation results into human-understandable text description output, meet the timeliness requirements of trend message generation, and provide technical support for military and civilian fields such as urban planning and regional / target monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0016] Figure 1 The flowchart of the method for generating a trend message according to an embodiment of the present disclosure is schematically shown.

[0017] Figure 2 The flowchart of preprocessing satellite time-series images according to an embodiment of the present disclosure is schematically shown.

[0018] Figure 3 The flowchart of extracting image features from satellite time-series images according to an embodiment of the present disclosure is schematically shown.

[0019] Figure 4 The flowchart schematically shows a flow chart of inputting context vectors into a t-way long short-term memory network in sequence to generate corresponding text description sentences according to an embodiment of the present disclosure.

[0020] Figure 5 The flowchart of generating a trend message based on reordered text description sentences according to an embodiment of the present disclosure is schematically shown.

[0021] Figure 6 The structure block diagram of the device for generating a movement message based on satellite time-series images according to an embodiment of the present disclosure is schematically shown; and

[0022] Figure 7 A block diagram of an electronic device according to a method for generating a trend message according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0023] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0024] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0025] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0026] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0027] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, disclosure and application of the data involved (including but not limited to user personal information) comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0028] Dynamic monitoring of typical regional changes and automatic generation of trend messages are the basis for intelligent interpretation of key areas / targets. Satellite observation is a large-scale and comprehensive means of earth observation. In particular, with the significant improvement in the spatial resolution and temporal resolution of space-based remote sensing satellites, the application efficiency of satellite observation technology in military and civilian fields such as key area observation and tracking of ground object evolution has become increasingly prominent. Satellite time series images refer to a collection of multiple observation sequence remote sensing images of the same area acquired within a time range. Trend messages refer to textual descriptions of changes in targets of interest in a given area. The generation of trend messages based on satellite time series images can characterize the trend change characteristics of regions / targets from a temporal dimension, providing an important way for trend research and judgment such as regional / target change pattern discovery and change trend prediction. The relevant results have important application value in the fields of national defense and people's livelihood.

[0029] To solve the above problems, embodiments of the present disclosure provide a method, apparatus and device for generating a trend message.

[0030] Figure 1 The flowchart of the method for generating a trend message according to an embodiment of the present disclosure is schematically shown.

[0031] like Figure 1As shown, the trend message generation method of this embodiment includes operations S110 to S230.

[0032] In operation S110 , satellite time-series images are acquired.

[0033] In this embodiment, the satellite time-series image includes multiple images. The Python installation package gdal is used to extract the name, shooting time, longitude and latitude information of each image in the satellite time-series image, and the information is associated and stored with the image entity.

[0034] In operation S120 , image features are extracted from the satellite time-series images.

[0035] In this embodiment, a multi-channel twin coding neural network is used to extract image features of satellite time series images, and the image features extracted by the multi-channel twin coding neural network are cascaded to characterize the changes in ground objects between satellite time series images. In some other embodiments, other neural networks can also be used as long as they can meet the extraction of image features, which is not limited here.

[0036] In operation S130, text description sentences corresponding to the image features are generated, and the text description sentences are reordered.

[0037] In this embodiment, a decoder neural network is used to understand and compile the extracted image features to generate text description sentences corresponding to the image features for easy understanding. In some other embodiments, other neural networks may also be used as long as they can satisfy the conversion of image features to text description sentences, which is not limited here.

[0038] In operation S140, a movement message is generated based on the reordered text description sentences.

[0039] In this embodiment, the dynamic message mainly includes a change description statement and whether the extracted image features have changed, etc. Specifically, the dynamic message is generated according to a message-specific format, and the generated message is template-matched and output.

[0040] Figure 2 The flowchart of preprocessing satellite time-series images according to an embodiment of the present disclosure is schematically shown.

[0041] Specifically, before acquiring the satellite time-series images in the above operation S110, the following steps may be included:

[0042] In operation S210, the satellite time-series images are pre-processed; wherein the satellite time-series images include image names and image header information.

[0043] In operation S220, image capturing times of satellite time-series images are extracted from image names.

[0044] In operation S230, the longitude and latitude of the image shooting location of the satellite time-series image are extracted from the image header information.

[0045] In this embodiment, the image names of satellite time-series images are parsed. The image names are stored in a 12-digit format, where the 1st to 4th digits represent the year of image shooting, the 5th to 6th digits represent the month of image shooting, the 7th to 8th digits represent the image shooting date, and the 9th to 12th digits represent the image shooting clock and minute, thereby extracting the specific image shooting time of the satellite time-series image, and the extracted image shooting time is associated and stored with the corresponding satellite time-series image for subsequent use.

[0046] In this embodiment, the image header information of the satellite time-series image is parsed. The format of the satellite time-series image is .tiff format, and the .tiff format file is parsed to extract the longitude and latitude information of the image shooting location. Specifically, the Python installation package gdal is used to parse the image header information, thereby extracting the longitude and latitude of the image shooting location of the satellite time-series image, and the extracted longitude and latitude of the image shooting location are associated with the corresponding satellite time-series image and stored for subsequent use.

[0047] Figure 3 The flowchart of extracting image features from satellite time-series images according to an embodiment of the present disclosure is schematically shown.

[0048] Specifically, extracting image features from satellite time-series images in the above operation S120 may include the following steps:

[0049] In operation S121, a feature vector corresponding to each satellite time-series image is obtained.

[0050] In this embodiment, a multi-channel twin encoding neural network is constructed to extract image features. The Conv5_3 layer in the VGG16 network model is used as the output layer to extract image features to retain the rich detail information of the ground objects in the remote sensing image.

[0051] Specifically, the characteristic dimension of the VGG16 network model used in the embodiment of the present disclosure is 512, which specifically includes: network structure, VGG16 has a total of 16 layers, 13 convolutional layers and 3 fully connected layers, 5 pooling layers; convolutional layers, a total of 13 convolutional layers, in order: 2 two-dimensional convolutions with a convolution kernel size of 3 and an output channel number of 64, 2 two-dimensional convolutions with a convolution kernel size of 3 and an output channel number of 128, 2 two-dimensional convolutions with a convolution kernel size of 3 and an output channel number of 128 and 1 two-dimensional convolution with a convolution kernel size of 1 and an output channel number of 128, 2 convolution kernel sizes are 3 output channels. The first layer is a two-dimensional convolution with a kernel size of 512 and a two-dimensional convolution with an output channel number of 512, and the second layer is a two-dimensional convolution with a kernel size of 1 and an output channel number of 512. There are two convolution layers with a kernel size of 3 and an output channel number of 512, and one two-dimensional convolution with a kernel size of 1 and an output channel number of 512; the fully connected layer, the last three layers are fully connected layers, including two fully connected layers with an output channel number of 4096 and a softmax layer; the pooling layers are located after the second convolution layer with 64 cores, the second convolution layer with 128 cores, the third convolution layer with 256 cores, the third convolution layer with 512 cores, and the sixth convolution layer with 512 cores.

[0052] In some other embodiments of the present disclosure, other neural network models may be used for extraction, such as VGGNet, GoogleNet, and ResNet models, which are not limited here.

[0053] In this embodiment, the satellite time-series images are input into the VGG16 network model to extract image features and complete the semantic representation of the target in the satellite time-series images. t ∈R H ×W×c, (t=1, 2,...,k), I t is a satellite time series image, C is the number of channels of the input satellite time series image, which can be 1 (such as SAR) or 3 (such as optical), h and W are the height and width of the satellite time series image respectively; the feature vector corresponding to each satellite time series image is expressed as f t (t=1, 2, ..., k).

[0054] In operation S122, two consecutive feature vectors are subtracted to generate a feature vector difference.

[0055] In this embodiment, there are k satellite time-series images, and the image features of two consecutive satellite time-series images are subtracted. The generated feature vector difference has k-1 groups, which are recorded as the encoded image features Fv t ={Fv t 1 , Fv t 2 , ...Fv t d}(t=1,2,...,k-1).

[0056] In operation S123, the feature vector difference is input into a long short-term memory network to obtain encoded image features.

[0057] In this embodiment, the attention mechanism is used to calculate the attention mask α, and the Fv t The context vector Fc of the satellite time series image is obtained by weighting t .

[0058] Specifically, the generating of the text description sentence corresponding to the image feature in the above operation S130 further includes the following operations:

[0059] The context vector Fc t The images are input into the t-way long short-term memory network in sequence, and the decoder is used to understand the input image features and generate the corresponding text description sentences. Specifically, the t-way long short-term memory network (LSTM) is used to predict the words at each decoding moment one by one, and the complete text description sentence is obtained by splicing.

[0060] Figure 4 The flowchart schematically shows a flow chart of inputting context vectors into a t-way long short-term memory network in sequence to generate corresponding text description sentences according to an embodiment of the present disclosure.

[0061] Specifically, the above-mentioned inputting the context vectors into the t-way long short-term memory network in sequence, using the decoder to understand the input image features, and generating the corresponding text description sentence may specifically include the following steps:

[0062] In operation S131, the context vector corresponding to the i-th moment is input into the word vector model to obtain the corresponding word vector.

[0063] In this embodiment, taking the t-way long short-term memory network at the i-th moment as an example, the word vector model is used to input the word w (i) Process and get the word vector Fw t (i) , let the dimension of the word vector be the same as the context vector Fc t (i) The dimensions are consistent.

[0064] In operation S132, the context vector corresponding to the i-th moment, the word vector corresponding to the context vector, and the hidden state vector corresponding to the i-1-th moment are input into a t-way long short-term memory network.

[0065] In operation S133, a hidden state vector corresponding to the i-th moment is generated.

[0066] In this embodiment, take x t (i) =[Fct (i) , Fw t (i) ], as the input of the t-way long short-term memory network, and the input x t (i) and the hidden state vector h at the i-1th moment t (i-1) Feed into the t-way long short-term memory network to generate the hidden state vector ht at the current moment (i) .

[0067] Specifically, Fw t (i) =embed(w (i) );h t (i) =LSTM(x t (i) ,h t (i-1) ). (i) represents a word in the training text at the i-th moment, embed(·) is the word vector model, and x t (i) is the concatenated LSTM input at the i-th moment, and LSTM(·) represents the LSTM network and parameter set.

[0068] In operation S134, the hidden state vector at the i-th moment is calculated using a normalized exponential function to obtain the probability distribution of the predicted word at the i-th moment.

[0069] In this embodiment, the hidden state vector is multiplied by the output matrix to obtain a feature vector with the same dimension as the dictionary capacity, and then the probability distribution of the predicted word at the i-th moment (p(y t (i) Specifically, (p(y t (i) )=softmax[E0(h t (i) )], where (p(y t (i) ) represents the probability distribution of the predicted word at the i-th moment, h t (i) represents the hidden state vector and E0 represents the output matrix.

[0070] In operation S135 , the probability distribution at the i-th moment is calculated using a cross entropy function to obtain a loss between the probability distribution and the actual label.

[0071] In this embodiment, the formula for calculating the loss between the probability distribution and the actual label is:

[0072]

[0073] Among them, loss t (i) represents the cross entropy loss at the i-th moment, D represents the dictionary capacity, (I, S) represents the input image sentence pair, and w ii is the weight.

[0074] In operation S136 , the loss is used to perform update training.

[0075] In this embodiment, the LSTM is updated and trained using loss.

[0076] In operation S137, the word corresponding to the maximum probability index position in the probability distribution is used as the predicted word at the i-th moment.

[0077] In operation S138, the words at all times are concatenated to form a complete sentence.

[0078] In this embodiment, the complete sentence Y obtained by splicing the predicted words in the t-th way t =[y t (0) ,y t (1) , ..., y t (i) ,END]; At the same time, the weight parameters of the LSTM model are:

[0079] θ * =argmax∑log(p(Y t |I,θ))

[0080] Among them, Y t It represents the complete sentence obtained by concatenating the predicted words in the t-th way, and I represents the “difference” between two consecutive satellite time series images input.

[0081] Figure 5 The flowchart of generating a trend message based on reordered text description sentences according to an embodiment of the present disclosure is schematically shown.

[0082] Specifically, in the above operation S140, generating a trend message based on the reordered text description sentences may include the following steps:

[0083] In operation S141, a movement message is generated according to a message-specific format.

[0084] In this embodiment, the captured time of the extracted images is sorted in chronological order, and the number of image observations is obtained, which is the total number of input satellite time series images. tThe information on whether there is any change is extracted to meet the timeliness requirement of trend message generation.

[0085] Specifically, the earliest shooting time and the latest shooting time of the satellite time series image are obtained; the number of image observations is obtained, which is the total number of input satellite time series images; the shooting longitude and latitude of the satellite time series image are obtained; the statements output by each LSTM are sorted according to the time sequence of the images; and it is determined whether the statements output by all t LSTMs contain a "happened" to obtain whether the time series image has changed.

[0086] In operation S142, the trend message is output according to the template, and the information extracted in S151 and the statements output by each LSTM are sorted according to the following template.

[0087] In this embodiment, the template adopts: (XX / XX / XXXX) to (XX / XX / XXXX) , (X times) Observe, discover (X longitude and latitude) ( Yes / No ) changes; (X place) of( position ) position, there has been a (slight / significant / general / …) change; among them, ( object ) becomes ( object )". In some other embodiments of the present disclosure, different output templates can be set according to actual application scenarios and change granularity, which is not limited here.

[0088] Based on the above-mentioned method for generating a trend message, the present disclosure also provides a device for generating a trend message based on satellite time-series images. Figure 6 The device is described in detail.

[0089] Figure 6 The structural block diagram of a device for generating a movement message based on satellite time-series images according to an embodiment of the present disclosure is schematically shown.

[0090] like Figure 6 As shown, the device 600 for generating a trend message based on satellite time-series images in this embodiment includes a collection module 610 , an extraction module 620 , an image text description sentence generation module 630 , a sorting module 640 and a trend message generation module 650 .

[0091] The acquisition module 610 is used to acquire satellite time-series images; the extraction module 620 is used to extract image features of satellite time-series images from the satellite time-series images; the image text description sentence generation module 630 is used to generate text description sentences corresponding to the image features; the sorting module 640 is used to reorder the text description sentences; the trend message generation module 650 is used to generate a trend message based on the reordered text description sentences.

[0092] In one embodiment, the acquisition module 610 can be used to perform the operation S110 described above, which will not be described in detail here. In one embodiment, the extraction module 620 can be used to perform the operation S120 described above, which will not be described in detail here. In one embodiment, the image text description sentence generation module 630 can be used to perform the operation S130 described above, which will not be described in detail here. In one embodiment, the sorting module 640 can be used to perform the operation S140 described above, which will not be described in detail here. In one embodiment, the trend message generation module 650 can be used to perform the operation S150 described above, which will not be described in detail here.

[0093] According to an embodiment of the present disclosure, any multiple modules among the acquisition module 610, the extraction module 620, the image text description sentence generation module 630, the sorting module 640, and the trend message generation module 650 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the acquisition module 610, the extraction module 620, the image text description sentence generation module 630, the sorting module 640, and the trend message generation module 650 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware or in a suitable combination of any of them. Alternatively, at least one of the acquisition module 610, the extraction module 620, the image text description sentence generation module 630, the sorting module 640 and the trend message generation module 650 can be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function can be executed.

[0094] The present disclosure also provides an electronic device, including: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above method.

[0095] Figure 7 A block diagram of an electronic device according to a method for generating a trend message according to an embodiment of the present disclosure is schematically shown.

[0096] like Figure 7As shown, the electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 707 to a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include an onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0097] In RAM 703, various programs and data required for the operation of electronic device 700 are stored. Processor 701, ROM 702 and RAM 703 are connected to each other via bus 704. Processor 701 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 702 and / or RAM 703. It should be noted that the program can also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0098] According to an embodiment of the present disclosure, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to the bus 704. The electronic device 700 may further include one or more of the following components connected to the I / O interface 705: an input portion 706 including a keyboard, a mouse, etc.; an output portion 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 708 including a hard disk, etc.; and a communication portion 709 including a network interface card such as a LAN card, a modem, etc. The communication portion 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage portion 708 as needed.

[0099] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0100] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 702 and / or RAM 703 described above and / or one or more memories other than ROM 702 and RAM 703.

[0101] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for generating a trend message, characterized in that: include: Acquire satellite time-series images; Extracting image features from the satellite time-series images; Generating text description sentences corresponding to the image features, and reordering the text description sentences; A movement message is generated based on the reordered text description sentences.

2. The method for generating a trend message according to claim 1, characterized in that: Before extracting the image features of the satellite time-series images from the satellite time-series images, the method further includes: Preprocessing the satellite time-series images; wherein the satellite time-series images include image names and image header information; Extracting the image capturing time of the satellite time-series image from the image name; The longitude and latitude of the location where the satellite time-series image is taken are extracted from the image header information.

3. The method for generating a trend message according to claim 1, characterized in that: The step of extracting the image features of the satellite time-series images from the satellite time-series images comprises: Obtaining a feature vector corresponding to each of the satellite time-series images; Subtracting two consecutive feature vectors to generate a feature vector difference; The feature vector difference is input into a long short-term memory network to obtain the encoded image features.

4. The method for generating a trend message according to claim 3, characterized in that: The encoded image features obtained by inputting the feature vector difference into the long short-term memory network include: Weighting the image features to obtain a context vector of the satellite time series image; The context vectors are sequentially input into a t-way long short-term memory network to generate corresponding text description sentences.

5. The method for generating a trend message according to claim 4, characterized in that: The step of sequentially inputting the context vectors of the satellite time series images into a t-way long short-term memory network to generate corresponding text description sentences comprises: Input the context vector corresponding to the i-th moment into the word vector model to obtain the corresponding word vector; Input the context vector corresponding to the i-th moment, the word vector corresponding to the context vector, and the hidden state vector corresponding to the i-1-th moment into the t-way long short-term memory network; Generate the hidden state vector corresponding to the i-th moment.

6. The method for generating a trend message according to claim 5, characterized in that: After generating the hidden state vector corresponding to the i-th moment, the method further comprises: Using a normalized exponential function to calculate the hidden state vector at the i-th moment, the probability distribution of the predicted word at the i-th moment is obtained; Using the cross entropy function to calculate the probability distribution at the i-th moment to obtain the loss between the probability distribution and the actual label; performing update training using the loss; The word corresponding to the maximum probability index position in the probability distribution is used as the predicted word at the i-th moment; Concatenate the words at all times to form a complete sentence.

7. The method for generating a trend message according to claim 1, characterized in that: The generating of the movement message based on the reordered text description sentences comprises: Generate a movement message according to a special message format; The movement message is output according to the template.

8. A method and device for generating a trend message, comprising: Acquisition module, used to obtain satellite time-series images; An extraction module, used for extracting image features of the satellite time-series images from the satellite time-series images; An image text description sentence generation module, used to generate a text description sentence corresponding to the image features; A sorting module, used for re-sorting the text description sentences; A movement message generating module is used to generate a movement message based on the reordered text description sentences.

9. An electronic device, comprising: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method according to any one of claims 1 to 7.

Images