Automatic screening method and system based on remote sensing image content

Abstract

The invention discloses an automatic screening method and system based on remote sensing image content, and relates to the technical field of satellite remote sensing image screening. By using a clustering algorithm to perform cluster analysis on regional parameters, K regional cluster clusters are obtained. For each regional cluster cluster, a remote sensing image screening model for judging whether a satellite image is selected and an image sequence number prediction model for evaluating the sorting sequence number of each satellite image are trained. A task area cluster cluster corresponding to a task area parameter is obtained by calculation. The remote sensing image screening model corresponding to the task area cluster cluster is used to screen out a set of task selected images. The image sequence number prediction model corresponding to the task area cluster cluster is used to number the satellite images. The automatic screening and sorting of satellite images is realized, the labor cost of screening personnel is reduced, and the screening efficiency of satellite remote sensing images is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite remote sensing image screening, and in particular to an automatic screening method and system based on remote sensing image content. Background Art

[0002] With the development of remote sensing technology, satellite images have become an important means for us to obtain information about the earth. However, due to the huge number of satellite images and the differences in image quality, how to effectively screen and process these images has become an important research topic;

[0003] The number of satellite images is very large, and the images collected by these satellites contain rich information about the earth, such as topography, landforms, vegetation, water bodies, etc.

[0004] However, due to the large number of satellites, the quality of satellite images varies. Due to differences in satellite sensor type, orbital altitude, shooting time and other factors, satellite images of the same area may have obvious quality differences; for example, some images may have low resolution and cannot clearly see the details of the ground; some images may be affected by clouds, atmospheric refraction, etc., resulting in distortion of image contrast and color;

[0005] Therefore, when checking satellite images of an area, it is necessary to screen the satellite images so as to select the satellite images with higher quality or more recent time;

[0006] Currently, the screening of satellite images is mainly done manually, which requires a lot of manpower or some customized screening rules, such as giving priority to those with less cloud cover, or selecting those with a more recent time when the cloud cover is the same. This screening rule is relatively rigid and often screens out satellite images of poor quality;

[0007] To this end, the present invention proposes an automatic screening method and system based on remote sensing image content. Summary of the invention

[0008] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an automatic screening method and system based on remote sensing image content, which realizes automatic screening and sorting of satellite images, reduces the labor cost of screening personnel, and improves the screening efficiency of satellite remote sensing images.

[0009] To achieve the above object, an automatic screening method based on remote sensing image content is proposed according to Embodiment 1 of the present invention, comprising the following steps:

[0010] Step 1: Collect several regional image datasets, and divide each regional image dataset into a selected dataset and an unselected dataset;

[0011] Step 2: Collect the regional parameters corresponding to each regional image dataset, and collect the image parameters of each satellite image in each selected dataset and unselected dataset;

[0012] Step 3: Use clustering algorithm to perform cluster analysis on regional parameters to obtain K regional clusters; K is the preset number of clusters;

[0013] Step 4: For each regional cluster, use the image parameters of each satellite image in the selected data set and the unselected data set to train a remote sensing image screening model to determine whether the satellite image is selected;

[0014] Step 5: Collect the display serial number of each satellite image in each selected dataset in each regional cluster;

[0015] Step 6: For each regional cluster, based on the image parameters and display sequence of each satellite image in each selected data set, an image sequence number prediction model is trained to evaluate the ranking sequence of each satellite image;

[0016] Step 7: After the remote sensing image screening model and the image sequence number prediction model are trained, the task area parameters and task image set corresponding to the latest image screening task are collected;

[0017] Step 8: Calculate and obtain the regional cluster corresponding to the task area parameters as the task area cluster;

[0018] Step 9: Use the remote sensing image screening model corresponding to the task area cluster to screen out the task selected image set from the task image set;

[0019] Step 10: Use the image sequence number prediction model corresponding to the task area cluster to number the satellite images in the task selected image set, and display the satellite images in the task selected image set according to the sequence of numbers;

[0020] The method of collecting a plurality of regional image data sets and dividing each regional image data set into a selected data set and an unselected data set is:

[0021] Each time the satellite images are manually screened, all satellite images corresponding to the screening are collected; all satellite images in each manual screening of satellite images are taken as a set of regional image data sets;

[0022] The satellite images selected by the screening personnel during each manual screening constitute the selected data set corresponding to the image data set of the region, and the other satellite images in the image data set of the region except the selected data set constitute the unselected data set;

[0023] The method of collecting the regional parameters corresponding to each regional image data set is:

[0024] For each regional image dataset, the median longitude, median latitude, longitude span, latitude span, region type and region area of ​​the ground region corresponding to the regional image dataset are collected during manual screening to form the regional parameters;

[0025] The method of collecting the image parameters of each satellite image in each selected data set and unselected data set is:

[0026] For each satellite image in each selected dataset and unselected dataset, the effective percentage of coverage area, resolution, cloud cover, side viewing angle, shooting time difference and ground feature resolution corresponding to the satellite image are collected as image parameters;

[0027] The effective percentage of coverage area is the proportion of the target area covered by the satellite image;

[0028] The calculation method of the effective percentage of the coverage area is as follows:

[0029] Collect the vector range of the target area (selection range);

[0030] The boundaries of the satellite image are converted into corresponding ground space coordinates using the RPC model, and the converted ground space coordinates constitute the image coverage area;

[0031] Calculate the range of the image coverage area within the vector range of the target area as the image effective coverage area, calculate the ratio of the image effective coverage area to the target area, and use the ratio as the effective percentage of the image coverage area;

[0032] The method of using the clustering algorithm to perform cluster analysis on the regional parameters and obtain K regional clusters is as follows:

[0033] The regional parameters corresponding to each regional image data set are formed into a one-dimensional array, and the one-dimensional array is used as the coordinates of a data point;

[0034] Use clustering algorithm to divide all data points into K clusters;

[0035] The method of training the remote sensing image screening model for determining whether a satellite image is selected is:

[0036] For each regional cluster:

[0037] The selection label of each satellite image in the selected dataset of all regional image datasets in the regional cluster is marked as 1, and the selection label of each satellite image in the unselected dataset is marked as 0;

[0038] The image parameters of each satellite image are combined into a feature vector as the input of a remote sensing image screening model. The remote sensing image screening model uses the selected label prediction value of each satellite image as the output, the selected label prediction value is 0 or 1, the selected label of each satellite image is used as the prediction target, the square of the difference between the selected label prediction value and the selected label is used as the first prediction error corresponding to each satellite image, and minimizing the sum of the first prediction errors of all satellite images is used as the training target; the remote sensing image screening model is trained until the sum of the first prediction errors reaches convergence and the training is stopped;

[0039] The method of collecting the display serial number of each satellite image in each selected data set in each regional cluster is as follows:

[0040] Each time the screening personnel manually screen the satellite images, for each satellite image in the selected data set, the serial number sorted by the screening personnel for the selected satellite image is collected as the display serial number;

[0041] The method of training the image sequence number prediction model for evaluating the ranking sequence number of each satellite image is as follows:

[0042] The image parameters of each satellite image are combined into a feature vector as the input of an image sequence number prediction model. The image sequence number prediction model outputs a display sequence number prediction value for each satellite image. The display sequence number prediction value is a positive integer. The display sequence number of each satellite image is used as a prediction target. The square of the difference between the display sequence number prediction value and the display sequence number is used as the second prediction error corresponding to each satellite image. The image sequence number prediction model takes minimizing the sum of the second prediction errors of all satellite images as a training target. The image sequence number prediction model is trained until the sum of the second prediction errors reaches convergence and the training is stopped.

[0043] The method of filtering out the task selected image set from the task image set is:

[0044] Input each satellite image in the task image set into the remote sensing image screening model, obtain the selection label prediction value output by the remote sensing image screening model, and save the satellite image with the selection label prediction value of 1 into the task selected image set;

[0045] The method of numbering the satellite images in the task selected image set is:

[0046] Each satellite image in the task selected image set is input into the image sequence number prediction model to obtain the display sequence number prediction value output by the image sequence number prediction model, and each satellite image is numbered in ascending order according to the display sequence number prediction value.

[0047] According to the second embodiment of the present invention, an automatic screening system based on remote sensing image content is proposed, which includes a sample data collection module, a model training module and an image screening module; wherein each module is electrically connected;

[0048] The sample data collection module is mainly used to collect several regional image data sets, and divide each regional image data set into a selected data set and an unselected data set, and collect the regional parameters corresponding to each regional image data set, and collect the image parameters of each satellite image in each selected data set and unselected data set, and collect the display serial number of each satellite image in each selected data set in each regional cluster; the sample data collection module sends the selected data set, the unselected data set, the regional parameters, the image parameters and the display serial number to the model training module;

[0049] The model training module is mainly used to perform cluster analysis on regional parameters using a clustering algorithm to obtain K regional clusters; K is the preset number of clusters. For each regional cluster, the image parameters of each satellite image in the selected data set and the unselected data set are used to train a remote sensing image screening model for determining whether a satellite image is selected. For each regional cluster, based on the image parameters and display serial number of each satellite image in each selected data set, an image serial number prediction model for evaluating the sorting serial number of each satellite image is trained; and the remote sensing image screening model and the image serial number prediction model of each regional cluster are sent to the image screening module;

[0050] Among them, the image screening module is mainly used to collect the task area parameters and task image set corresponding to the latest image screening task after the training of the remote sensing image screening model and the image sequence number prediction model is completed, calculate the regional clustering cluster corresponding to the task area parameters as the task area clustering cluster, use the remote sensing image screening model corresponding to the task area clustering cluster to screen out the task selected image set from the task image set, use the image sequence number prediction model corresponding to the task area clustering cluster to number the satellite images in the task selected image set, and display the satellite images in the task selected image set according to the numbering order.

[0051] An electronic device according to Embodiment 3 of the present invention includes: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0052] The processor executes the above-mentioned automatic screening method based on remote sensing image content by calling the computer program stored in the memory.

[0053] A computer-readable storage medium according to embodiment 4 of the present invention stores a rewritable computer program;

[0054] When the computer program is executed on a computer device, the computer device is enabled to execute the above-mentioned automatic screening method based on remote sensing image content.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] The present invention collects several regional image data sets in advance, divides each regional image data set into a selected data set and an unselected data set, collects regional parameters corresponding to each regional image data set, collects image parameters of each satellite image in each selected data set and the unselected data set, uses a clustering algorithm to perform cluster analysis on the regional parameters, and obtains K regional clustering clusters; K is the number of preset clustering clusters, for each regional clustering cluster, uses the image parameters of each satellite image in the selected data set and the unselected data set to train a remote sensing image screening model for judging whether a satellite image is selected, collects the display serial number of each satellite image in each selected data set in each regional clustering cluster, and for each regional clustering cluster, based on the image parameters and display serial number of each satellite image in each selected data set, trains an image serial number pre-training for evaluating the sorting serial number of each satellite image. After the remote sensing image screening model and the image sequence number prediction model are trained, the task area parameters and task image sets corresponding to the latest image screening task are collected, and the regional clustering clusters corresponding to the task area parameters are calculated as the task area clustering clusters. The remote sensing image screening model corresponding to the task area clustering clusters is used to screen out the task selected image set from the task image set. The image sequence number prediction model corresponding to the task area clustering clusters is used to number the satellite images in the task selected image set, and the satellite images in the task selected image set are displayed according to the numbering sequence. The remote sensing image screening model is used to learn the screening strategy of the screening personnel, and the image sequence number prediction model is used to learn the sorting strategy of the screened screening personnel, thereby realizing automatic screening and sorting of satellite images, reducing the labor cost of screening personnel and improving the screening efficiency of satellite remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a flow chart of an automatic screening method based on remote sensing image content in Embodiment 1 of the present invention;

[0058] Figure 2 It is a module connection relationship diagram of the automatic screening system based on remote sensing image content in Example 2 of the present invention;

[0059] Figure 3 This is a schematic diagram of the structure of an electronic device in Embodiment 3 of the present invention;

[0060] Figure 4 This is a schematic diagram of the computer-readable storage medium structure in Example 4 of the present invention. DETAILED DESCRIPTION

[0061] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] Example 1

[0063] like Figure 1 As shown, the automatic screening method based on remote sensing image content includes the following steps:

[0064] Step 1: Collect several regional image datasets, and divide each regional image dataset into a selected dataset and an unselected dataset;

[0065] Step 2: Collect the regional parameters corresponding to each regional image dataset, and collect the image parameters of each satellite image in each selected dataset and unselected dataset;

[0066] Step 3: Use clustering algorithm to perform cluster analysis on regional parameters to obtain K regional clusters; K is the preset number of clusters;

[0067] Step 4: For each regional cluster, use the image parameters of each satellite image in the selected data set and the unselected data set to train a remote sensing image screening model to determine whether the satellite image is selected;

[0068] Step 5: Collect the display serial number of each satellite image in each selected dataset in each regional cluster;

[0069] Step 6: For each regional cluster, based on the image parameters and display sequence of each satellite image in each selected data set, an image sequence number prediction model is trained to evaluate the ranking sequence of each satellite image;

[0070] Step 7: After the remote sensing image screening model and the image sequence number prediction model are trained, the task area parameters and task image set corresponding to the latest image screening task are collected;

[0071] Step 8: Calculate and obtain the regional cluster corresponding to the task area parameters as the task area cluster;

[0072] Step 9: Use the remote sensing image screening model corresponding to the task area cluster to screen out the task selected image set from the task image set;

[0073] Step 10: Use the image sequence number prediction model corresponding to the task area cluster to number the satellite images in the task selected image set, and display the satellite images in the task selected image set according to the sequence of numbers;

[0074] The method of collecting a plurality of regional image data sets and dividing each regional image data set into a selected data set and an unselected data set is as follows:

[0075] Each time the satellite images are manually screened, all satellite images corresponding to the screening are collected; all satellite images in each manual screening of satellite images are taken as a set of regional image data sets;

[0076] It should be noted that each time during manual screening, the screening personnel select a ground area, for example, the XX city area, and then screen out some satellite images that meet the expected rules from all satellite images of the area, and sort the selected satellite images; because the expected rules for the satellites photographing the ground area include but are not limited to whether they meet the cloud range, side viewing angle range, resolution range, ground feature recognition, etc.; and the method of sorting the selected satellite images can be based on the above-mentioned expected rules, manually sorting the quality of the satellite images; the cloud range is set because the clouds will block the ground, thereby causing the ground in the satellite image to be blocked, affecting the viewing experience; the side viewing angle range is set because the side viewing angle will affect the field of view covered by the satellite image, for example, when the side viewing angle is small, that is, when looking directly at the ground, the covered field of view is the most comprehensive; the resolution range is set because the resolution will affect the clarity of the satellite image; the ground feature recognition is set because it is necessary to clearly distinguish the various ground features in the satellite image, for example, the color difference of ground objects on snowy days is small, which will make the objects difficult to distinguish;;

[0077] The satellite images selected by the screening personnel during each manual screening constitute the selected data set corresponding to the image data set of the region, and the other satellite images in the image data set of the region except the selected data set constitute the unselected data set;

[0078] Furthermore, the method of collecting the regional parameters corresponding to each regional image data set is:

[0079] For each regional image dataset, the median longitude, median latitude, longitude span, latitude span, region type and region area of ​​the ground region corresponding to the regional image dataset are collected during manual screening to form the regional parameters;

[0080] It should be noted that the median longitude and the median latitude are respectively the center values ​​of the longitude range covered by the ground area, and the median latitude is the center value of the latitude range covered by the ground area. The area type can be the topographic type of the ground area or the regional construction type;

[0081] Furthermore, the method of collecting the image parameters of each satellite image in each selected data set and unselected data set is:

[0082] For each satellite image in each selected dataset and unselected dataset, the effective percentage of coverage area, resolution, cloud cover, side viewing angle, shooting time difference and ground feature resolution corresponding to the satellite image are collected as image parameters;

[0083] It should be noted that the calculation method of the effective percentage of the coverage area is:

[0084] Collect the vector range of the target area (selection range);

[0085] The boundaries of the satellite image are converted into corresponding ground space coordinates using the RPC model, and the converted ground space coordinates constitute the image coverage area;

[0086] Calculate the range of the image coverage area within the vector range of the target area as the image effective coverage area, calculate the ratio of the image effective coverage area to the target area, and use the ratio as the effective percentage of the image coverage area;

[0087] The resolution and side angle of view are parameters directly calculated by the satellite according to the position of the satellite in orbit when taking pictures, and are inherent properties of satellite images;

[0088] The shooting time difference refers to the difference between the shooting time and the screening time;

[0089] The cloud amount calculation method is a conventional technical means in the field. For example, the Chinese patent with the authorization announcement number CN107529647B discloses a cloud amount calculation method for cloud images based on a multi-layer unsupervised sparse learning network. First, a forward layer-by-layer sparse automatic encoder is used to perform unsupervised layer-by-layer feature encoding on the image to obtain high-order semantic information. Then, the cloud image is divided into thick clouds, thin clouds and clear sky using the high-order semantic information. Finally, the total cloud amount in the cloud image is calculated using the "spatial correlation method". The present invention will not be repeated here.

[0090] The object resolution is expressed by using the variance of all pixel values ​​in the satellite image;

[0091] Furthermore, the method of using a clustering algorithm to perform cluster analysis on regional parameters to obtain K regional clusters is as follows:

[0092] The regional parameters corresponding to each regional image data set are formed into a one-dimensional array, and the one-dimensional array is used as the coordinates of a data point;

[0093] Using a clustering algorithm to divide all data points into K clusters; specifically, the clustering algorithm can be any one of the K-means or FCM algorithms;

[0094] Furthermore, for each regional cluster, the image parameters of each satellite image in the selected data set and the unselected data set are used to train a remote sensing image screening model for determining whether a satellite image is selected:

[0095] For each regional cluster:

[0096] The selection label of each satellite image in the selected dataset of all regional image datasets in the regional cluster is marked as 1, and the selection label of each satellite image in the unselected dataset is marked as 0;

[0097] The image parameters of each satellite image are used to form a feature vector as the input of a remote sensing image screening model. The remote sensing image screening model uses the selection label prediction value of each satellite image as an output, and the selection label prediction value is 0 or 1. The selection label of each satellite image is used as a prediction target, and the square of the difference between the selection label prediction value and the selection label is used as the first prediction error corresponding to each satellite image. Minimizing the sum of the first prediction errors of all satellite images is used as a training target; the remote sensing image screening model is trained until the sum of the first prediction errors reaches convergence, and the training is stopped to train a remote sensing image screening model that predicts whether the satellite image is selected based on the image parameters of the satellite image; the remote sensing image screening model is any one of the classification models such as a deep neural network model (DNN) or a deep belief network (DBN); it can be understood that the sum of the first prediction errors is a mean square error;

[0098] Furthermore, the method of collecting the display serial number of each satellite image in each selected data set in each regional cluster is as follows:

[0099] Each time the screener manually screens the satellite images, for each satellite image in the selected data set, the serial number sorted by the screener for the selected satellite image is collected as the display serial number; it can be understood that the display serial number reflects the screener's recognition of the quality of the satellite image to a certain extent, and only the satellite image with the best quality can be ranked first, so the display serial number can also be used to express the quality of the satellite image;

[0100] Furthermore, for each regional cluster, based on the image parameters and display sequence number of each satellite image in each selected data set, the image sequence number prediction model for evaluating the sorting sequence number of each satellite image is trained as follows:

[0101] The image parameters of each satellite image are used to form a feature vector as the input of the image sequence number prediction model. The image sequence number prediction model takes the display sequence number prediction value of each satellite image as the output. The display sequence number prediction value is a positive integer. The display sequence number of each satellite image is used as the prediction target. The square of the difference between the display sequence number prediction value and the display sequence number is used as the second prediction error corresponding to each satellite image. The image sequence number prediction model takes minimizing the sum of the second prediction errors of all satellite images as the training target. The image sequence number prediction model is trained until the sum of the second prediction errors reaches convergence, and the training is stopped to train an image sequence number prediction model that predicts the display sequence number of the satellite image according to the image parameters of the satellite image. The remote sensing image screening model is any one of the regression models such as the polynomial regression model or the support vector machine regression model (SVR). It can be understood that the sum of the second prediction errors is the mean square error.

[0102] Furthermore, the task area parameter is an area parameter of a ground area where the satellite image is located and needs to be screened for the latest image screening task;

[0103] The task image set is all satellite images that need to be screened for the latest image screening task;

[0104] The method of calculating and obtaining the regional cluster corresponding to the task area parameter as the task area cluster is as follows:

[0105] Calculate the distance between the data points composed of the task area parameters and the center point of each regional cluster, and take the regional cluster with the closest distance as the task area cluster;

[0106] The method of filtering out the task selected image set from the task image set is:

[0107] Input each satellite image in the task image set into the remote sensing image screening model, obtain the selection label prediction value output by the remote sensing image screening model, and save the satellite image with the selection label prediction value of 1 into the task selected image set;

[0108] The method of numbering the satellite images in the task selected image set is:

[0109] Each satellite image in the task selected image set is input into the image sequence number prediction model to obtain the display sequence number prediction value output by the image sequence number prediction model, and each satellite image is numbered in ascending order according to the display sequence number prediction value.

[0110] Example 2

[0111] like Figure 2As shown, the automatic screening system based on remote sensing image content includes a sample data collection module, a model training module and an image screening module; wherein each module is electrically connected;

[0112] The sample data collection module is mainly used to collect several regional image data sets, and divide each regional image data set into a selected data set and an unselected data set, and collect the regional parameters corresponding to each regional image data set, and collect the image parameters of each satellite image in each selected data set and unselected data set, and collect the display serial number of each satellite image in each selected data set in each regional cluster; the sample data collection module sends the selected data set, the unselected data set, the regional parameters, the image parameters and the display serial number to the model training module;

[0113] The model training module is mainly used to perform cluster analysis on regional parameters using a clustering algorithm to obtain K regional clusters; K is the preset number of clusters. For each regional cluster, the image parameters of each satellite image in the selected data set and the unselected data set are used to train a remote sensing image screening model for determining whether a satellite image is selected. For each regional cluster, based on the image parameters and display serial number of each satellite image in each selected data set, an image serial number prediction model for evaluating the sorting serial number of each satellite image is trained; and the remote sensing image screening model and the image serial number prediction model of each regional cluster are sent to the image screening module;

[0114] Among them, the image screening module is mainly used to collect the task area parameters and task image set corresponding to the latest image screening task after the training of the remote sensing image screening model and the image sequence number prediction model is completed, calculate the regional clustering cluster corresponding to the task area parameters as the task area clustering cluster, use the remote sensing image screening model corresponding to the task area clustering cluster to screen out the task selected image set from the task image set, use the image sequence number prediction model corresponding to the task area clustering cluster to number the satellite images in the task selected image set, and display the satellite images in the task selected image set according to the numbering order.

[0115] Example 3

[0116] Figure 3 1 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 3 As shown, according to another aspect of the present application, an electronic device 100 is also provided. The electronic device 100 may include one or more processors and one or more memories, and the processor may include a CPU 102 and a GPU 109. The memory stores a computer readable code, and when the computer readable code is executed by one or more processors, the automatic screening method based on remote sensing image content as described above may be executed.

[0117] The method or system according to the embodiment of the present application can also be used by Figure 3 The electronic device architecture shown in FIG. Figure 3 As shown, the electronic device 100 may include a bus 101, one or more CPUs 102, a read-only memory (ROM) 103, a random access memory (RAM) 104, a communication port 105 connected to a network, an input / output component 106, a hard disk 107, a GPU 109, etc. The storage device in the electronic device 100, such as the ROM 103 or the hard disk 107, may store the automatic screening method based on remote sensing image content provided in the present application. The automatic screening method based on remote sensing image content may, for example, include the following steps: Step 1: Collect several regional image data sets, and divide each regional image data set into a selected data set and an unselected data set; Step 2: Collect regional parameters corresponding to each regional image data set, and collect image parameters of each satellite image in each selected data set and unselected data set; Step 3: Use a clustering algorithm to perform cluster analysis on the regional parameters to obtain K regional clustering clusters; K is the preset number of clustering clusters; Step 4: For each regional clustering cluster, use the image parameters of each satellite image in the selected data set and the unselected data set to train a remote sensing image screening model for determining whether the satellite image is selected; Step 5: Collect the display serial number of each satellite image in each selected data set in each regional clustering cluster; Step 6: Step 6: For each regional clustering cluster, based on the image parameters and display serial number of each satellite image in each selected data set, train an image serial number prediction model to evaluate the sorting serial number of each satellite image; Step 7: After the remote sensing image screening model and the image serial number prediction model are trained, collect the task area parameters and task image set corresponding to the latest image screening task; Step 8: Calculate and obtain the regional clustering cluster corresponding to the task area parameters as the task area clustering cluster; Step 9: Use the remote sensing image screening model corresponding to the task area clustering cluster to filter out the task selected image set from the task image set; Step 10: Use the image serial number prediction model corresponding to the task area clustering cluster to number the satellite images in the task selected image set, and display the satellite images in the task selected image set according to the numbering order;

[0118] Furthermore, the electronic device 100 may further include a user interface 108. Figure 3 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 3 One or more components of an electronic device are shown.

[0119] Example 4

[0120] Figure 4 Schematic diagram of a computer-readable storage medium structure provided by an embodiment of the present application. Figure 4 As shown, a computer-readable storage medium 200 according to an embodiment of the present application is shown. Computer-readable instructions are stored on the computer-readable storage medium 200. When the computer-readable instructions are executed by the processor, the automatic screening method based on remote sensing image content according to the embodiment of the present application described with reference to the above figures can be executed. The computer-readable storage medium 200 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0121] In addition, according to the implementation of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium stores machine-readable instructions, the machine-readable instructions can be run by a processor to execute instructions corresponding to the method steps provided by the present application, and when the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0122] The methods, apparatuses, and devices of the present application may be implemented in many ways. For example, the methods, apparatuses, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0123] The above preset parameters or preset thresholds are all set by technicians in this field according to actual conditions or obtained through large-scale data simulation.

[0124] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An automatic screening method based on remote sensing image content, characterized in that: The following steps are involved: Collecting a number of regional image data sets, and dividing each regional image data set into a selected data set and an unselected data set; Collect the regional parameters corresponding to each regional image dataset, and collect the image parameters of each satellite image in each selected dataset and unselected dataset; Use clustering algorithm to perform cluster analysis on regional parameters to obtain K regional clusters; K is the preset number of clusters; For each regional cluster, the image parameters of each satellite image in the selected and unselected datasets are used to train a remote sensing image screening model to determine whether the satellite image is selected. Collect the display serial number of each satellite image in each selected dataset in each regional cluster; For each regional cluster, an image number prediction model is trained to evaluate the ranking number of each satellite image based on the image parameters and display number of each satellite image in each selected data set; After the remote sensing image screening model and the image sequence number prediction model are trained, the task area parameters and task image sets corresponding to the latest image screening task are collected; Calculate and obtain the regional clustering cluster corresponding to the task area parameters as the task area clustering cluster; Use the remote sensing image screening model corresponding to the task area cluster to screen out the task selected image set from the task image set; The satellite images in the task selected image set are numbered using the image sequence number prediction model corresponding to the task area clustering cluster, and the satellite images in the task selected image set are displayed in the order of the numbers.

2. The automatic screening method based on remote sensing image content according to claim 1 is characterized in that: The method of collecting a plurality of regional image data sets and dividing each regional image data set into a selected data set and an unselected data set is: Each time the satellite images are manually screened, all satellite images corresponding to the screening are collected; all satellite images in each manual screening of satellite images are taken as a set of regional image data sets; During each manual screening, some satellite images screened by the screening personnel constitute a selected data set corresponding to the image data set of the region, and other satellite images in the image data set of the region except the selected data set constitute an unselected data set.

3. The automatic screening method based on remote sensing image content according to claim 2 is characterized in that: The method of collecting the regional parameters corresponding to each regional image data set is: For each regional image dataset, the median longitude, median latitude, longitude span, latitude span, region type and region area of ​​the ground area corresponding to the regional image dataset are collected during manual screening to form the regional parameters.

4. The automatic screening method based on remote sensing image content according to claim 3 is characterized in that: The method of collecting the image parameters of each satellite image in each selected data set and unselected data set is: For each satellite image in each selected and unselected dataset, the effective percentage of coverage area, resolution, cloud cover, side viewing angle, shooting time difference and ground feature resolution corresponding to each satellite image are collected as image parameters.

5. The automatic screening method based on remote sensing image content according to claim 4 is characterized in that: The method of training the remote sensing image screening model for determining whether a satellite image is selected is: For each regional cluster: The selection label of each satellite image in the selected dataset of all regional image datasets in the regional cluster is marked as 1, and the selection label of each satellite image in the unselected dataset is marked as 0; The image parameters of each satellite image are combined into a feature vector as the input of the remote sensing image screening model. The remote sensing image screening model uses the selected label prediction value of each satellite image as the output. The selected label prediction value is 0 or 1. The selected label of each satellite image is used as the prediction target. The square of the difference between the selected label prediction value and the selected label is used as the first prediction error corresponding to each satellite image. Minimizing the sum of the first prediction errors of all satellite images is used as the training target. The remote sensing image screening model is trained until the sum of the first prediction errors reaches convergence.

6. The automatic screening method based on remote sensing image content according to claim 5 is characterized in that: The method of training the image sequence number prediction model for evaluating the ranking sequence number of each satellite image is as follows: The image parameters of each satellite image are combined into a feature vector as the input of an image sequence number prediction model. The image sequence number prediction model outputs a display sequence number prediction value for each satellite image. The display sequence number prediction value is a positive integer. The display sequence number of each satellite image is used as a prediction target. The square of the difference between the display sequence number prediction value and the display sequence number is used as the second prediction error corresponding to each satellite image. The image sequence number prediction model takes minimizing the sum of the second prediction errors of all satellite images as a training target. The image sequence number prediction model is trained until the sum of the second prediction errors reaches convergence, and the training is stopped.

7. The automatic screening method based on remote sensing image content according to claim 6 is characterized in that: The method of filtering out the task selected image set from the task image set is: Input each satellite image in the task image set into the remote sensing image screening model, obtain the selection label prediction value output by the remote sensing image screening model, and save the satellite image with the selection label prediction value of 1 into the task selected image set; The method of numbering the satellite images in the task selected image set is: Each satellite image in the task selected image set is input into the image sequence number prediction model to obtain the display sequence number prediction value output by the image sequence number prediction model, and each satellite image is numbered in ascending order according to the display sequence number prediction value.

8. An automatic screening system based on remote sensing image content, which is used to implement the automatic screening method based on remote sensing image content as described in any one of claims 1 to 7, characterized in that: It includes a sample data collection module, a model training module and an image screening module; wherein each module is electrically connected to each other; The sample data collection module is used to collect several regional image data sets, and divide each regional image data set into a selected data set and an unselected data set, and collect the regional parameters corresponding to each regional image data set, and collect the image parameters of each satellite image in each selected data set and unselected data set, and collect the display serial number of each satellite image in each selected data set in each regional cluster; the sample data collection module sends the selected data set, the unselected data set, the regional parameters, the image parameters and the display serial number to the model training module; The model training module is used to perform cluster analysis on regional parameters using a clustering algorithm to obtain K regional clusters; K is the preset number of clusters. For each regional cluster, the image parameters of each satellite image in the selected data set and the unselected data set are used to train a remote sensing image screening model for determining whether a satellite image is selected. For each regional cluster, based on the image parameters and display serial number of each satellite image in each selected data set, an image serial number prediction model for evaluating the sorting serial number of each satellite image is trained; and the remote sensing image screening model and the image serial number prediction model of each regional cluster are sent to the image screening module; Among them, the image screening module is used to collect the task area parameters and task image set corresponding to the latest image screening task after the training of the remote sensing image screening model and the image sequence number prediction model is completed, calculate the regional cluster cluster corresponding to the task area parameters as the task area cluster cluster, use the remote sensing image screening model corresponding to the task area cluster cluster, filter out the task selected image set from the task image set, use the image sequence number prediction model corresponding to the task area cluster cluster, number the satellite images in the task selected image set, and display the satellite images in the task selected image set according to the numbering order.

9. An electronic device, characterized in that: include: processor and memory, wherein The memory stores a computer program that can be called by the processor; The processor executes the automatic screening method based on remote sensing image content as described in any one of claims 1 to 7 in the background by calling the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: A rewritable computer program is stored thereon; When the computer program is executed on a computer device, the computer device is enabled to execute the automatic screening method based on remote sensing image content as described in any one of claims 1 to 7.

Images