Temporal Remote Sensing Image Segmentation Method, Device, Electronic Device and Storage Medium

Abstract

The present invention provides a method, apparatus, electronic device and storage medium for segmenting time-series remote sensing images. The method for segmenting time-series remote sensing images includes: obtaining a target time-series remote sensing image to be segmented; performing multi-dimensional feature analysis on each pixel point in the target time-series image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between the multi-dimensional features in the multi-dimensional feature sequence, processing each pixel point in the target time-series image to obtain a target segmentation spatio-temporal cube; using the target segmentation spatio-temporal cube to segment the target time-series image to obtain a time-series image segmentation result. The present invention realizes the automatic segmentation of time-series remote sensing images, which is a new extension of the object-oriented idea in the spatio-temporal domain.

Description

Technical Field

[0001] The present invention relates to the technical field of processing of time-series remote sensing images, and in particular, to a method, device, electronic device and storage medium for segmenting time-series remote sensing images. Background Art

[0002] A time-series remote sensing image, also known as a satellite image time series, refers to a set of image data obtained by a satellite sensor for the same geographical area over a continuous period of time. At present, the development of remote sensing technology has greatly improved the spatial and temporal resolutions of the obtained time-series image data, forming consistent and repetitive observations, which provides support for the monitoring and evaluation of surface cover and its change information. Therefore, time-series image data has been widely applied in fields such as land use / land cover, agricultural change, resource investigation, and environmental monitoring. The purpose of time-series image segmentation is to determine and generate the basic units for time-series image analysis applications, and the spatio-temporal objects obtained by segmentation will be used as the raw materials for feature extraction, classification mapping, and change analysis of time-series images.

[0003] In the prior art, the pixel and object models and their segmentation methods used in time-series image analysis usually process time and space separately. For example, pixel time-series segmentation focuses on extracting the time-varying information of ground objects; while change detection focuses on extracting the spatial change information of ground objects. Therefore, the remote sensing image segmentation methods in the prior art have the problem that they cannot effectively achieve time-series image segmentation. Summary of the Invention

[0004] The present invention provides a method, device, electronic device and storage medium for segmenting time-series remote sensing images, which are used to solve the defect that the prior art is not applicable to time-series remote sensing image segmentation, and realize accurate and effective segmentation of time-series remote sensing images.

[0005] The present invention provides a method for segmenting time-series remote sensing images, including: obtaining a target time-series remote sensing image to be segmented; performing multi-dimensional feature analysis on each pixel point in the target time-series remote sensing image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, processing each pixel point in the target time-series remote sensing image to obtain a target segmentation spatio-temporal cube sequence; using the target segmentation spatio-temporal cube sequence to segment the target time-series remote sensing image to obtain a time-series remote sensing image segmentation result.

[0006] A method for segmenting temporal remote sensing images provided by the present invention, processing each pixel point in the target temporal remote sensing image based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, to obtain a target segmentation spatio-temporal cube sequence, includes: based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, for each pixel point in the target temporal remote sensing image, obtaining an intermediate spatio-temporal cube; in the intermediate spatio-temporal cube, determining a candidate spatio-temporal cube; based on the multi-dimensional feature matching degree between the candidate spatio-temporal cube and a comparison spatio-temporal cube, processing the candidate spatio-temporal cube to obtain a target segmentation spatio-temporal cube sequence; wherein, the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube.

[0007] A method for segmenting temporal remote sensing images provided by the present invention, the multi-dimensional feature matching degree includes a spatial dimension matching degree, and processing each pixel point in the target temporal remote sensing image based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence to obtain a target segmentation spatio-temporal cube sequence includes: obtaining a first spectral feature matching degree and a shape feature matching degree in the spatial dimension matching degree; performing a weighted sum of the first spectral feature matching degree and the shape feature matching degree to obtain a first target matching degree; based on the first target matching degree, for each pixel point in the target temporal remote sensing image, obtaining a target segmentation spatio-temporal cube sequence.

[0008] A method for segmenting temporal remote sensing images provided by the present invention, based on the first target matching degree, for each pixel point in the target temporal remote sensing image, obtaining a target segmentation spatio-temporal cube includes: obtaining a pixel point sequence corresponding to the target temporal remote sensing image; the pixel point sequence is a sequence composed of each pixel point in the target temporal remote sensing image; in the pixel point sequence, determining a candidate pixel point as a candidate spatio-temporal cube; when the first target matching degree is greater than a preset matching degree threshold, merging the candidate spatio-temporal cube with a comparison spatio-temporal cube to obtain a merged spatio-temporal cube, forming an initial spatio-temporal cube; the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube; determining the merged spatio-temporal cube as the candidate spatio-temporal cube, and executing the step of when the first target matching degree is greater than a preset matching degree threshold, merging the candidate spatio-temporal cube with a comparison spatio-temporal cube to obtain a merged spatio-temporal cube.

[0009] A method for segmenting time-series remote sensing images provided by the present invention, the process of obtaining the shape feature matching degree includes: obtaining a first compactness value and a first smoothness value corresponding to the candidate spatio-temporal cube, and obtaining a second compactness value and a second smoothness value corresponding to the comparison spatio-temporal cube; and obtaining a third compactness value and a third smoothness value of the merged spatio-temporal cube; based on the first compactness value, the first smoothness value, the second compactness value, the second smoothness value, the third compactness value and the third smoothness value, obtaining a compactness difference value and a smoothness difference value; performing weighted summation on the compactness difference value and the smoothness difference value to obtain a shape difference value; based on the corresponding relationship between the shape difference value and the shape feature matching degree, obtaining the shape feature matching degree.

[0010] A method for segmenting time-series remote sensing images provided by the present invention, the process of obtaining the first spectral feature matching degree includes: obtaining the number of spectral bands in the target time-series remote sensing image; based on the number of spectral bands, obtaining the first spectral feature matching degree.

[0011] A method for segmenting time-series remote sensing images provided by the present invention, the multi-dimensional feature matching degree includes a time dimension matching degree, and the process of processing each pixel point in the target time-series remote sensing image based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence to obtain a target segmentation spatio-temporal cube sequence includes: obtaining a second spectral feature matching degree in the time dimension matching degree; based on the second spectral feature matching degree, processing each pixel point in the target time-series remote sensing image to obtain a target segmentation spatio-temporal cube sequence.

[0012] The present invention also provides a device for segmenting time-series remote sensing images, including: a target time-series remote sensing image acquisition module for acquiring a target time-series remote sensing image to be segmented; a multi-dimensional feature sequence obtaining module for performing multi-dimensional feature analysis on each pixel point in the target time-series remote sensing image to obtain a multi-dimensional feature sequence; a target segmentation spatio-temporal cube obtaining module for processing each pixel point in the target time-series remote sensing image based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence to obtain a target segmentation spatio-temporal cube; and a segmented time-series remote sensing image obtaining module for segmenting the target time-series remote sensing image by using the target segmentation spatio-temporal cube to obtain a segmented time-series remote sensing image segmentation result.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the steps of any one of the above-mentioned time-series remote sensing image segmentation methods are implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above-described temporal remote sensing image segmentation methods are implemented.

[0015] The temporal remote sensing image segmentation method, device, electronic device and storage medium provided by the present invention obtain a target temporal remote sensing image to be segmented; perform multi-dimensional feature analysis on each pixel point in the target temporal remote sensing image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, process each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube sequence; use the target segmentation spatio-temporal cube sequence to segment the target temporal remote sensing image to obtain a segmented temporal remote sensing image. Based on the relationship between the above multi-dimensional features, a target segmentation spatio-temporal cube sequence is obtained, so as to accurately obtain the segmented temporal remote sensing image. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 is one of the scene schematic diagrams of the temporal remote sensing image segmentation method provided by the present invention;

[0018] Figure 2 is the second flow schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0019] Figure 3 is the third flow schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0020] Figure 4 is the fourth flow schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0021] Figure 5 is the fifth flow schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0022] Figure 6 is the sixth flow schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0023] Figure 7 is the seventh flow schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0024] Figure 8It is the eighth flow schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0025] Figure 9 It is one of the structural schematic diagrams of the temporal remote sensing image segmentation method provided by the present invention;

[0026] Figure 10 It is the second structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0027] Figure 11 It is one of the effect schematic diagrams of the temporal remote sensing image segmentation method provided by the present invention;

[0028] Figure 12 It is the third structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0029] Figure 13 It is the fourth structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0030] Figure 14 It is the fifth structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0031] Figure 15 It is the sixth structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0032] Figure 16 It is the seventh structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0033] Figure 17 It is the eighth structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0034] Figure 18 It is the ninth structural schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0035] Figure 19 It is the second effect schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0036] Figure 20 It is the third effect schematic diagram of the temporal remote sensing image segmentation method provided by the present invention;

[0037] Figure 21 It is the structural schematic diagram of the temporal remote sensing image segmentation device provided by the present invention;

[0038] Figure 22 It is the structural schematic diagram of the electronic device provided by the present invention. Specific Embodiments

[0039] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] The method for segmenting temporal remote sensing images provided by the present invention can be applied to an application environment as Figure 1 shown, and is specifically applied to a temporal remote sensing image segmentation system. The temporal remote sensing image segmentation system includes a terminal 102 and a server 104. Among them, the terminal 102 communicates with the server 104 through a network. The server 104 executes a method for segmenting temporal remote sensing images. Specifically, the server 104 obtains the target temporal remote sensing image to be segmented from the terminal 102; performs multi-dimensional feature analysis on each pixel point in the target temporal remote sensing image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between the multi-dimensional features in the multi-dimensional feature sequence, processes each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube; uses the target segmentation spatio-temporal cube to segment the target temporal remote sensing image to obtain a segmented temporal remote sensing image. Among them, the terminal 102 can be, but is not limited to, various temporal remote sensing image acquisition devices, cameras, personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices, and the server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0041] The following combines Figures 2 - 8 to describe the method for segmenting temporal remote sensing images of the present invention.

[0042] In one embodiment, as Figure 2 shown, a method for segmenting temporal remote sensing images is provided. Taking the method applied to the Figure 1 server as an example, it includes the following steps:

[0043] Step 202, obtain the target temporal remote sensing image to be segmented.

[0044] Specifically, the server can obtain the target temporal remote sensing image to be segmented for temporal remote sensing image segmentation directly or indirectly. The temporal remote sensing image stored locally or transmitted to the terminal in real time can be obtained as the target temporal remote sensing image to be segmented.

[0045] In one embodiment, when the server receives a time-series remote sensing image processing instruction, the time-series remote sensing image identification of the target time-series remote sensing image to be segmented is carried in the time-series remote sensing image processing instruction. Through this time-series remote sensing image identification, the server can obtain the target time-series remote sensing image to be segmented from the time-series remote sensing images stored locally.

[0046] In one embodiment, the target time-series remote sensing image to be segmented can be acquired by a time-series remote sensing image acquisition device. The time-series remote sensing image acquisition device is connected to the server. When the time-series remote sensing image acquisition device receives the instruction from the server to obtain a time-series remote sensing image, it transmits the acquired real-time time-series remote sensing image or the time-series remote sensing image stored locally in the time-series remote sensing image acquisition device to the server. The server uses the received time-series remote sensing image as the target time-series remote sensing image to be segmented for time-series remote sensing image processing. The time-series remote sensing image acquisition device includes various cameras, scanners, various cameras, time-series remote sensing image acquisition cards, etc.

[0047] In one embodiment, the time-series remote sensing image acquisition device can transmit the acquired time-series remote sensing images to the server sequentially or in batches at a certain time interval. After the server obtains the time-series remote sensing images, it can store them locally for future use, or perform real-time processing after receiving the time-series remote sensing images.

[0048] Step 204: Perform multi-dimensional feature analysis on each pixel point in the target time-series remote sensing image to obtain a multi-dimensional feature sequence.

[0049] Among them, the multi-dimensional feature refers to the time-series remote sensing image features in multiple dimensions. For example, the time-series remote sensing image features in the time dimension or the time-series remote sensing image features in the space dimension, etc. The above-mentioned time-series remote sensing image features refer to the iconic features that can reflect the difference between one time-series remote sensing image and another. For example, if the time-series remote sensing image feature of time-series remote sensing image A is three spectral bands and the time-series remote sensing image feature of time-series remote sensing image B is four spectral bands, then the spectral band can be used as the iconic feature to distinguish time-series remote sensing image A from time-series remote sensing image B.

[0050] Specifically, after the server obtains the target time-series remote sensing image, the arrangement of each pixel point in the target time-series remote sensing image has a certain order. After performing multi-dimensional analysis on each pixel point in the target time-series remote sensing image, according to the arrangement of each pixel point in the target time-series remote sensing image, the multi-dimensional features are arranged in the same way, so that each pixel point corresponds one-to-one with its corresponding multi-dimensional feature, thereby obtaining a multi-dimensional feature sequence. The direction of this sequence can be at least one of vertical or horizontal.

[0051] Step 206: Based on the multi-dimensional feature matching degrees among the multi-dimensional features in the multi-dimensional feature sequence, process each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube sequence.

[0052] Among them, the multi-dimensional feature matching degree refers to the similarity degree between multi-dimensional features; the higher the similarity degree, the higher the matching degree; the lower the similarity degree, the lower the matching degree.

[0053] Specifically, after the server obtains the multi-dimensional feature sequence, it calculates the multi-dimensional feature matching degree between the multi-dimensional features corresponding to each spatio-temporal cube in the target temporal remote sensing image and the multi-dimensional features corresponding to the spatio-temporal cubes in the neighborhood of this spatio-temporal cube, and for each pixel point in the target temporal remote sensing image, it obtains the target pixel segmentation spatio-temporal cube.

[0054] In one embodiment, a pixel point is selected from each pixel point in the target temporal remote sensing image as a candidate spatio-temporal cube, the multi-dimensional features of this candidate spatio-temporal cube are obtained, and the multi-dimensional features corresponding to the spatio-temporal cubes in the neighborhood of this candidate spatio-temporal cube are obtained. According to the multi-dimensional feature matching degree between the multi-dimensional features of this candidate spatio-temporal cube and the multi-dimensional features corresponding to the spatio-temporal cubes in the neighborhood of this candidate spatio-temporal cube, for each pixel point in the target temporal remote sensing image, the target pixel segmentation spatio-temporal cube is obtained.

[0055] In one embodiment, when the multi-dimensional feature matching degree is less than or equal to the preset matching degree, the two spatio-temporal cubes corresponding to the two multi-dimensional features are not merged. Using the region growing algorithm, the spatio-temporal cubes in the neighborhood of the candidate spatio-temporal cube are used as new candidate spatio-temporal cubes to perform a new judgment of the multi-dimensional feature matching degree.

[0056] In one embodiment, when the multi-dimensional feature matching degree is greater than the preset matching degree, the two spatio-temporal cubes corresponding to the two multi-dimensional features are merged, and the merged spatio-temporal cube is used as a new spatio-temporal cube to perform a new judgment of the multi-dimensional feature matching degree.

[0057] In one embodiment, there is a negative correlation between the multi-dimensional feature matching degree and the multi-dimensional feature difference degree. The greater the multi-dimensional feature difference degree, the smaller the multi-dimensional feature matching degree; the smaller the multi-dimensional feature difference degree, the greater the multi-dimensional feature matching degree. The multi-dimensional feature matching degree can be determined through the multi-dimensional feature difference degree.

[0058] In one embodiment, the determination of the multi-dimensional feature difference degree can be determined through the difference degree between the spatio-temporal cubes in the time dimension or the spectral features of the spatio-temporal cubes, or can be determined through the difference degree between the spatio-temporal cubes in the space dimension or the difference degree between the shape features and spectral features of the spatio-temporal cubes.

[0059] Step 208: Use the target pixel segmentation spatio-temporal cube to segment the target temporal remote sensing image, and obtain the segmentation result of the temporal remote sensing image.

[0060] Specifically, after the server obtains the target pixel segmentation spatio-temporal cube, it uses the target pixel segmentation spatio-temporal cube to segment the target temporal remote sensing image, and obtains the segmented temporal remote sensing image.

[0061] In the above temporal remote sensing image segmentation method, by obtaining the target temporal remote sensing image to be segmented; performing multi-dimensional feature analysis on each pixel point in the target temporal remote sensing image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, for each pixel point in the target temporal remote sensing image, obtain the target pixel segmentation spatio-temporal cube; use the target pixel segmentation spatio-temporal cube to segment the target temporal remote sensing image to obtain the segmented temporal remote sensing image. Thus, the purpose of accurately obtaining the segmented temporal remote sensing image is achieved.

[0062] In one embodiment, as Figure 3 shown, based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, processing each pixel point in the target temporal remote sensing image to obtain the target pixel segmentation spatio-temporal cube sequence includes:

[0063] Step 302: Based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, for each pixel point in the target temporal remote sensing image, obtain an intermediate spatio-temporal cube sequence.

[0064] Among them, the intermediate spatio-temporal cube refers to a spatio-temporal cube composed of multiple pixel points with multi-dimensional features.

[0065] Specifically, after the server obtains the multi-dimensional feature matching degree, it merges two or more pixel points whose multi-dimensional feature matching degree reaches the multi-dimensional feature matching degree threshold to obtain a spatio-temporal cube. However, before judging the multi-dimensional feature matching degree between this spatio-temporal cube and the spatio-temporal cubes in its neighborhood, the obtained spatio-temporal cube can be used as the intermediate spatio-temporal cube for obtaining the target pixel segmentation spatio-temporal cube.

[0066] Step 304: In the intermediate spatio-temporal cube, determine the candidate spatio-temporal cube.

[0067] Specifically, after obtaining the intermediate spatio-temporal cube, any one can be selected as the candidate spatio-temporal cube.

[0068] In one embodiment, for the intermediate spatio-temporal cube with a candidate identifier in the intermediate spatio-temporal cube, it can be determined as the candidate spatio-temporal cube to improve the efficiency of temporal remote sensing image segmentation.

[0069] Step 306: Based on the multi-dimensional feature matching degree between the candidate spatio-temporal cube and the comparison spatio-temporal cube, obtain a target segmented spatio-temporal cube sequence for the candidate spatio-temporal cube; wherein, the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube.

[0070] Specifically, after determining the candidate spatio-temporal cube, the server merges spatio-temporal cubes through the multi-dimensional feature matching degree between the candidate spatio-temporal cube and the comparison spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube to obtain a target pixel-segmented spatio-temporal cube, traverses each candidate spatio-temporal cube, makes a judgment on the multi-dimensional feature matching degree, and obtains a target pixel-segmented spatio-temporal cube sequence.

[0071] In this embodiment, through the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, for each pixel point in the target temporal remote sensing image, an intermediate spatio-temporal cube is obtained. In the intermediate spatio-temporal cube, a candidate spatio-temporal cube is determined, and based on the multi-dimensional feature matching degree between the candidate spatio-temporal cube and the comparison spatio-temporal cube, a target pixel-segmented spatio-temporal cube is obtained for the candidate spatio-temporal cube. In this embodiment, the spatio-temporal cube can be used as the smallest comparison unit to determine the target pixel spatio-temporal cube, and the purpose of accurately and quickly determining the target pixel spatio-temporal cube can be achieved.

[0072] In one embodiment, as Figure 4 shown, the multi-dimensional feature matching degree includes a spatial dimension matching degree. Processing each pixel point in the target temporal remote sensing image based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, and obtaining a target pixel-segmented spatio-temporal cube sequence includes:

[0073] Step 402: Obtain the first spectral feature matching degree and shape feature matching degree in the spatial dimension matching degree.

[0074] Among them, the spectral feature refers to the feature composed of spectral information in the temporal remote sensing image data. For example, the spectral information of bands such as R, G, B, and NIR in the remote sensing temporal remote sensing image. The spectral feature matching degree refers to the similarity degree between spectral features. The higher the similarity degree, the higher the spectral feature matching degree; the lower the similarity degree, the lower the spectral feature matching degree. The shape feature refers to the feature that uses shape to distinguish different objects in the temporal remote sensing image. For example, for circular objects and rectangular features in the temporal remote sensing image, the circular and rectangular shapes can be used to distinguish circular objects and rectangular features.

[0075] Specifically, shape features can be obtained through methods such as the boundary feature method or the Fourier shape descriptor method, and spectral features can be extracted through spectral extraction methods such as the spectral correlation analysis method, the joint entropy analysis method, or the hybrid spectral analysis method. Then, the shape features of each spatio-temporal cube and the spectral features of each spatio-temporal cube extracted are compared respectively to obtain the first spectral feature matching degree and the shape feature matching degree.

[0076] In one embodiment, the first spectral feature heterogeneity difference and the heterogeneity difference of the shape feature can be utilized, and the negative correlation relationship between the first spectral feature heterogeneity difference and the first spectral feature matching degree, as well as the negative correlation relationship between the heterogeneity difference of the shape feature and the shape feature matching degree, can be used to determine the first spectral feature matching degree and the shape feature matching degree. The greater the first spectral feature heterogeneity difference, the smaller the first spectral feature matching degree; the smaller the first spectral feature heterogeneity difference, the greater the first spectral feature matching degree; the greater the heterogeneity difference of the shape feature, the smaller the shape feature matching degree; the smaller the heterogeneity difference of the shape feature, the greater the shape feature matching degree.

[0077] In one embodiment, taking the time-series image as an example, the first spectral feature heterogeneity difference is denoted as H color , H color measures the spectral difference between adjacent spatio-temporal cubes in the spatial dimension or before and after the merging of spatio-temporal cubes. The first spectral feature heterogeneity difference H color is expressed by the formula:

[0078]

[0079] where m represents the number of spectral bands of the time-series image participating in the segmentation, ω k is the weight value of the k-th band, m represents the total number of pixels contained in the bottom surface of the spatial domain of the spatio-temporal cube, ΔT is the time-span length, cb1 and cb2 respectively represent two adjacent spatio-temporal cubes in the spatial domain before merging, cb represents the spatio-temporal cube obtained by merging cb1 and cb2, and h k represents the spectral difference of the spatio-temporal cube on the k-th band, and the value range of k is from 1 to m.

[0080] Specifically, the spectral difference h k of the spatio-temporal cube on the k-th band can be expressed by the formula:

[0081]

[0082] T represents the number of time points contained in the spatio-temporal cube, represents the spectral standard deviation of the k-th band at the i-th time point.

[0083] Specifically, the spectral standard deviation of the k-th band at the i-th time point is expressed as the formula:

[0084]

[0085] where n represents the number of pixels in the spatial domain of the i-th time point of the spatio-temporal cube, and p j represents the spectral band value of the j-th pixel, and represents the average spectral band value of all n pixels.

[0086] In one embodiment, the shape heterogeneity difference is determined by spatio-temporal compactness and spatio-temporal smoothness, and the shape heterogeneity difference is denoted as H shape , the spatio-temporal compactness is denoted as h cpt , and the spatio-temporal smoothness is denoted as h smooth , and H shape reflects the shape difference before and after the merger of adjacent spatio-temporal cubes in the spatial domain. H shape can be expressed as the formula:

[0087] H shape = ω cpt ·h cpt +(1 - ω cpt )·h smooth (4)

[0088] where h cpt represents the compactness difference between spatio-temporal cubes before and after the merger, and h smooth is the smoothness difference, and ω cpt (0 ≤ ω cpt ≤ 1) represents the weight value of the compactness difference in the shape heterogeneity. By calculating the differences in compactness and smoothness before and after the merger of spatio-temporal solid objects, the change in shape heterogeneity caused by each merger can be determined.

[0089] Specifically, the compactness difference h cpt between spatio-temporal cubes can be expressed as the formula:

[0090] h cpt = n cb ·ΔT cb ·Cpt cb -(n cb1 ·ΔT cb1 ·Cpt cb1 + n cb2 ·ΔT cb2 ·Cpt cb2 )(5)

[0091] where Cpt represents the average compactness difference of a spatio-temporal cube over the entire spatio-temporal domain and can be expressed as the formula:

[0092]

[0093] Among them, l represents the perimeter of the polygon in the spatial domain of the spatio-temporal cube, and n is the number of pixels included in the spatial domain of the cube.

[0094] Specifically, the spatio-temporal smoothness h smooth is expressed as the formula:

[0095] h smooth = n cb ·ΔT cb ·Smooth cb -(n cb1 ·ΔT cb1 ·Smooth cb1 + n cb2 ·ΔT cb2 ·Smooth cb2 )(7)

[0096] Among them, Smooth represents the average smoothness of the spatio-temporal cube within the entire spatio-temporal range it contains, and can be expressed as the formula:

[0097]

[0098] Among them, b i represents the shortest possible side length value of the bounding box of the polygon in the spatial domain of the i-th time point image layer of the spatio-temporal cube, that is, the shortest side length value of the minimum bounding rectangle.

[0099] Step 404, perform a weighted sum of the first spectral feature matching degree and the shape feature matching degree to obtain the first target matching degree.

[0100] Specifically, after the server obtains the first spectral feature matching degree and the shape feature matching degree, it can perform a weighted sum of the first spectral feature matching degree and the shape feature matching degree to obtain the first target matching degree.

[0101] In one embodiment, based on the correspondence between the first spectral feature matching degree and the first spectral feature heterogeneity difference, and the correspondence between the shape feature matching degree and the shape heterogeneity difference, the first target difference can be obtained by performing a weighted sum of the first spectral feature heterogeneity difference and the shape heterogeneity difference, and the corresponding first target matching degree can be obtained through the first target difference.

[0102] In one embodiment, the first spectral feature heterogeneity difference is expressed as H color , the heterogeneity difference of the shape feature is expressed as H shape , the heterogeneity difference of the spatial dimension is expressed as S h , and the weight parameter of the heterogeneity of the shape feature is expressed as ω shape(0 ≤ ω shape ≤ 1), the heterogeneity difference S of the spatial dimension h is expressed as a formula:

[0103] S h = ω shape ·H shape + (1 - ω shape )·H color (9)

[0104] It can be understood that the weight parameter ω of the heterogeneity of the shape features shape is larger, the greater the influence of the heterogeneity difference of the shape features on the heterogeneity difference of the entire spatial dimension, and the smaller the influence of the heterogeneity difference of the first spectral feature on the heterogeneity difference of the entire spatial dimension; the weight parameter ω of the shape heterogeneity shape is smaller, the smaller the influence of the heterogeneity difference of the shape features on the heterogeneity difference of the entire spatial dimension, and the greater the influence of the heterogeneity difference of the first spectral feature on the heterogeneity difference of the entire spatial dimension.

[0105] Step 406: Based on the first target matching degree, process each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube sequence.

[0106] Among them, the target pixel spatio-temporal cube refers to a spatio-temporal cube or a set of spatio-temporal cubes with a certain spatio-temporal volume that can segment the target temporal remote sensing image. For example, a spatio-temporal cube composed of multiple spatio-temporal cubes, etc.

[0107] Specifically, after obtaining the first target matching degree, the server can determine whether to merge one or more of the spatio-temporal cubes in the target temporal remote sensing image according to the first target matching degree, so as to obtain the target segmentation spatio-temporal cube.

[0108] In this embodiment, by obtaining the first spectral feature matching degree and the shape feature matching degree in the spatial dimension matching degree, weighting and summing the first spectral feature matching degree and the shape feature matching degree to obtain the first target matching degree, and based on the first target matching degree, processing each pixel point in the target temporal remote sensing image to obtain the target segmentation spatio-temporal cube, the purpose of accurately obtaining the target segmentation spatio-temporal cube can be achieved.

[0109] In one embodiment, as Figure 5 shown, based on the first target matching degree, processing each pixel point in the target temporal remote sensing image to obtain the target segmentation spatio-temporal cube includes:

[0110] Step 502: Obtain the pixel point sequence corresponding to the target temporal remote sensing image; the pixel point sequence is a sequence composed of each pixel point in the target temporal remote sensing image.

[0111] Specifically, the server extracts the pixel points of the target temporal remote sensing image using a pixel point sequence extraction plugin, and arranges them in order according to the coordinate values of each pixel point to obtain the pixel point sequence corresponding to the target temporal remote sensing image; the pixel point sequence is a sequence composed of each pixel point in the target temporal remote sensing image.

[0112] Step 504, determine a candidate spatio-temporal cube in the pixel point sequence.

[0113] Specifically, after the server obtains the pixel point sequence corresponding to the target temporal remote sensing image, it selects any pixel point in the pixel point sequence as the candidate spatio-temporal cube; or, uses the pixel point set with a candidate identifier as the candidate spatio-temporal cube, etc.

[0114] Step 506, when the first target matching degree is greater than the preset matching degree threshold, merge the candidate spatio-temporal cube with the comparison spatio-temporal cube to obtain a merged spatio-temporal cube; the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube.

[0115] Among them, the preset matching degree threshold refers to the critical value of the matching degree. When it is greater than this critical value, it is considered that the first target matching degree meets the requirements of the matching degree. When it is less than or equal to this critical value, it is considered that the first target matching degree does not meet the requirements of the matching degree.

[0116] Specifically, when the first target matching degree is greater than the preset matching degree threshold, the first target matching degree between the candidate spatio-temporal cube and the comparison spatio-temporal cube meets the requirements of the matching degree, and the candidate spatio-temporal cube and the comparison spatio-temporal cube are merged to obtain a merged spatio-temporal cube.

[0117] Step 508, determine the merged spatio-temporal cube as the candidate spatio-temporal cube, and execute the step of merging the candidate spatio-temporal cube with the comparison spatio-temporal cube to obtain a merged spatio-temporal cube when the first target matching degree is greater than the preset matching degree threshold.

[0118] Specifically, after the server obtains the merged spatio-temporal cube, it determines the merged spatio-temporal cube as the new candidate spatio-temporal cube, and continues to use the new candidate spatio-temporal cube as the candidate spatio-temporal cube to compare the spatio-temporal cubes within the neighborhood.

[0119] In this embodiment, by obtaining the pixel point sequence corresponding to the target temporal remote sensing image; the pixel point sequence is a sequence composed of each pixel point in the target temporal remote sensing image, determining a candidate spatio-temporal cube in the pixel point sequence, and when the first target matching degree is greater than the preset matching degree threshold, merging the candidate spatio-temporal cube with the comparison spatio-temporal cube to obtain a merged spatio-temporal cube; the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube, the purpose of improving the efficiency of obtaining the target segmentation spatio-temporal cube can be achieved.

[0120] In one embodiment, as Figure 6 shown, the process of obtaining the shape feature matching degree includes:

[0121] Step 602, obtain the first compactness value and the first smoothness value corresponding to the candidate spatio-temporal cube, and obtain the second compactness value and the second smoothness value corresponding to the comparison spatio-temporal cube; and obtain the third compactness value and the third smoothness value of the merged spatio-temporal cube.

[0122] Specifically, the first compactness value is denoted as h cpt1 , the first smoothness value is denoted as h smooth1 , the second compactness value is denoted as h cpt2 , the second smoothness value is denoted as h smooth2 , the third compactness value is denoted as h cpt3 , the first smoothness value is denoted as h smooth3 , and the above values can be obtained through the average compactness of the spatio-temporal cube and the number of spatio-temporal cubes.

[0123] Step 604, based on the first compactness value, the first smoothness value, the second compactness value, the second smoothness value, the third compactness value and the third smoothness value, obtain the compactness difference value and the smoothness difference value.

[0124] Specifically, the compactness difference value is denoted as h cpt , the smoothness difference value is denoted as h smooth ;

[0125] The compactness difference value h cpt is expressed by the formula:

[0126] h cpt =h cpt3 -(h cpt1 +h cpt2 ) (10)

[0127] The smoothness difference value h smooth is expressed by the formula:

[0128] h smooth =h smooth3 -(h smooth1 +h smooth2 ) (11)

[0129] Wherein, h cpt3 is expressed by the formula:

[0130] h cpt3 =n cb ·ΔT cb ·Cpt cb (12)

[0131] hcpt1 Expressed as a formula:

[0132] h cpt1 =n cb1 ·ΔT cb1 ·Cpt cb1 (13)

[0133] h cpt2 Expressed as a formula:

[0134] h cpt2 =n cb2 ·ΔT cb2 ·Cpt cb2 ) (14)

[0135] Wherein, h smooth3 Expressed as a formula:

[0136] h smooth3 =n cb ·ΔT cb ·Smooth cb (15)

[0137] h smooth1 Expressed as a formula:

[0138] h smooth1 =n cb1 ·ΔT cb1 ·smooth cb1 (16)

[0139] h smooth2 Expressed as a formula:

[0140] h smooth2 =n cb2 ·ΔT cb2 ·Smooth cb2 (17)

[0141] Step 606, perform a weighted sum of the compactness difference value and the smoothness difference value to obtain a shape difference value.

[0142] Specifically, after the server obtains the compactness difference value and the smoothness difference value, according to the different weights corresponding to the compactness difference value and the smoothness difference value, perform a weighted sum of the compactness difference value and the smoothness difference value to obtain a shape difference value. The shape difference value is denoted as H shape , then the shape difference value H shape Expressed as a formula:

[0143] H shape =ω cpt ·h cpt +(1 - ω cpt )·h smooth(18)

[0144] Among them, ω cpt (0 ≤ ω cpt ≤ 1) represents the weight value of the compactness difference in the shape feature heterogeneity difference. The larger this weight value, the greater the influence of the compactness difference on the shape difference value, and the smaller the influence of the smoothness difference value on the shape difference value; the smaller this weight value, the smaller the influence of the compactness difference on the shape difference value, and the greater the influence of the smoothness difference value on the shape difference value.

[0145] Step 608: Obtain the shape feature matching degree based on the correspondence between the shape difference value and the shape feature matching degree.

[0146] Specifically, after the server obtains the shape difference value, it obtains the shape feature matching degree according to the correspondence between the shape difference value and the shape feature matching degree.

[0147] In one embodiment, there is a negative correlation between the shape difference value and the shape feature matching degree. The larger the shape difference value, the smaller the shape feature matching degree; the smaller the shape difference value, the larger the shape feature matching degree. After obtaining the shape difference value, the shape feature matching degree can be correspondingly obtained. It can be understood that the shape difference value can reflect the heterogeneity difference of the shape feature. The larger the shape difference value, the greater the heterogeneity difference of the shape feature; the smaller the shape difference value, the smaller the heterogeneity difference of the shape feature.

[0148] In this embodiment, by obtaining the first compactness value and the first smoothness value corresponding to the candidate spatio-temporal cube, and obtaining the second compactness value and the second smoothness value corresponding to the comparison spatio-temporal cube; and obtaining the third compactness value and the third smoothness value of the merged spatio-temporal cube, based on the first compactness value, the first smoothness value, the second compactness value, the second smoothness value, the third compactness value and the third smoothness value, obtaining the compactness difference value and the smoothness difference value, performing weighted summation on the compactness difference value and the smoothness difference value to obtain the shape difference value, and obtaining the shape feature matching degree based on the correspondence between the shape difference value and the shape feature matching degree, the purpose of accurately obtaining the shape feature matching degree can be achieved.

[0149] In one embodiment, as Figure 7 shown, the process of obtaining the first spectral feature matching degree includes:

[0150] Step 702: Obtain the number of spectral bands in the target temporal remote sensing image.

[0151] Specifically, by using the spectral band extraction plugin, the spectral bands in the target temporal remote sensing image are obtained, and then the number of spectral bands in the target temporal remote sensing image is obtained. For example, if the target temporal remote sensing image obtained is an RGB temporal remote sensing image, the number of spectral bands of the target temporal remote sensing image obtained by using the extraction plugin is three.

[0152] Step 704, based on the number of spectral bands, obtain the first spectral feature matching degree.

[0153] Specifically, after the server obtains the number of spectral bands, using the weight value ω of each spectral band k , obtain the first spectral feature heterogeneity difference. The number of spectral bands is denoted as m, and the first spectral feature heterogeneity difference is expressed as the formula:

[0154]

[0155] According to the relationship between the first spectral feature heterogeneity difference and the first spectral feature matching degree, obtain the first spectral feature matching degree.

[0156] In this embodiment, by obtaining the number of spectral bands in the target temporal remote sensing image and based on the number of spectral bands, obtaining the first spectral feature matching degree, the purpose of accurately obtaining the first spectral feature matching degree can be achieved.

[0157] In one embodiment, as Figure 8 shown, the multi-dimensional feature matching degree includes the time dimension matching degree. Based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, each pixel point in the target temporal remote sensing image is processed to obtain the target segmentation spatio-temporal cube sequence, including:

[0158] Step 802, obtain the second spectral feature matching degree in the time dimension matching degree.

[0159] Among them, the time dimension refers to the time domain dimension. For example, the temporal remote sensing image is a temporal remote sensing image with time domain characteristics.

[0160] Specifically, the second spectral feature matching degree can be obtained by obtaining the heterogeneity difference of the second spectral feature.

[0161] In one embodiment, the heterogeneity difference of the second spectral feature is denoted as T h , then the heterogeneity difference T h of the second spectral feature is expressed as the formula:

[0162]

[0163] Among them, H t refers to the average spectral heterogeneity difference of the spatio-temporal cube in the time domain, and is expressed as the formula:

[0164]

[0165] Among them, m represents the number of spectral bands of the time series data, and n represents the number of pixels included in the bottom polygon of the current spatio-temporal cube. As Figure 9 shown, represents the spectral standard deviation of the longitudinal vector of time pixels formed by the i-th spatio-temporal cube in the spatio-temporal cube space domain at all time points included in the cube on the k-th band, which is expressed by the formula:

[0166]

[0167] Specifically, due to the spatio-temporal coverage characteristics of the spatio-temporal cube, when obtaining the time heterogeneity difference of the spatio-temporal cube, it is necessary to obtain the average spectral heterogeneity of each spatio-temporal cube on the longitudinal vector of time pixels formed by all spatio-temporal cubes in the included spatial domain. Therefore, in this embodiment, the longitudinal vector of time pixels formed by expanding each pixel in the spatial domain of the spatio-temporal cube on the included time domain is used to sum the spectral standard deviations of each longitudinal vector of time pixels, and then the average is taken with respect to the number of pixels in the spatial domain of the spatio-temporal cube. Similarly, in order to ensure the relative consistency of the calculation of the time heterogeneity of different spatio-temporal cubes, the time heterogeneity H of each cube is weighted and calculated using the volume size of the spatio-temporal solid object. t

[0168] Step 804, based on the second spectral feature matching degree, for each pixel point in the target time series remote sensing image, obtain the target segmented spatio-temporal cube.

[0169] Specifically, after the server obtains the second spectral feature matching degree, based on this second spectral feature matching degree, the candidate spatio-temporal cube in the target time series remote sensing image can be merged with the spatio-temporal cubes in the neighborhood, or no processing is performed, and the spatio-temporal cubes in the neighborhood of the spatio-temporal cube are continuously searched as new candidate spatio-temporal cubes, and the step of obtaining the second spectral feature matching degree in the time dimension matching degree is executed to obtain the target segmented spatio-temporal cube.

[0170] In this embodiment, by obtaining the second spectral feature matching degree in the time dimension matching degree and based on the second spectral feature matching degree, for each pixel point in the target time series remote sensing image, the target segmented spatio-temporal cube is obtained, which can achieve the purpose of accurately and quickly obtaining the target segmented spatio-temporal cube in the time dimension.

[0171] ​In one embodiment, taking time-series remote sensing images as an example, time-series image segmentation is to aggregate adjacent pixels that are relatively homogeneous in the spatio-temporal domain to form a spatio-temporal cube. Therefore, time-series image segmentation is a process of merging spatio-temporal cubes in time-series images. For each pixel in the time-series image, they are gradually merged into spatio-temporal cubes. In each processing step, a pair of smaller cubes that are adjacent in time or space are considered whether to be merged into a larger cube, and whether to merge the spatio-temporal cubes is based on the local spatio-temporal heterogeneity criterion between the cubes, and this criterion measures the similarity of spatio-temporal characteristics between two adjacent spatio-temporal cubes. The overall goal of segmentation is to minimize the heterogeneity change in each merging process. Therefore, a spatio-temporal cube should be merged with the most suitable cube in the neighborhood. Spatio-temporal heterogeneity can not only be used to judge whether a pair of cubes should be merged, but also be used to measure the heterogeneity difference between adjacent spatio-temporal cubes before and after merging. When performing time-series image segmentation processing, for each pair of spatio-temporal cubes to be merged, first calculate the spatio-temporal heterogeneity difference value between the cubes. If it is less than the set maximum difference degree, then merge; if it is greater than or equal to the set maximum difference degree, then do not merge. When there is no heterogeneity difference between adjacent spatio-temporal cubes in the entire time-series image that is less than the given maximum difference degree, the image segmentation ends.

[0172] In one embodiment, taking time-series remote sensing images as an example, this image has continuity in both the spatial domain and the time domain. Then, for the formation of the corresponding spatio-temporal cube of this image, the spatial domain and the time domain can be combined to accurately depict the spatio-temporal continuity and distribution characteristics of geographical objects in the time-series image. A spatio-temporal cube refers to a spatio-temporal three-dimensional region composed of relatively homogeneous pixels within a certain continuous spatio-temporal domain in the time-series image. As Figure 10 shown in the schematic diagram of this spatio-temporal cube, the spatial domain of the time-series image is expressed by the X and Y axes, and the time domain is expressed by the T axis, forming a continuous spatio-temporal three-dimensional space. Therefore, a spatio-temporal cube C i (S i ,T i ) is a continuous and homogeneous spatio-temporal domain that contains both spatial information (S i ) and time information (T i ). Among them, S i is the spatial domain of C i . Within a spatio-temporal cube, S i has the same top surface and bottom surface. It can be understood that for the convenience of graphical display, regular-shaped top and bottom surfaces are used in this embodiment, but in fact, it can be an irregular-shaped polygon to reflect the actual spatial shape of the ground object. For different spatio-temporal cubes, they have different spatial domain distributions to facilitate the analysis of the spatial changes of ground objects. S i reflects the spatial shape and position information of the spatio-temporal cube. T i (Ti = <s i , e i >) is the time domain, which consists of the starting time point s i and the ending time point e i and reflects the continuous distribution information of the spatio-temporal cube in terms of time. Therefore, overall, the spatio-temporal cube C i can express that the ground objects are in a continuous homogeneous state within the spatial domain S i and the time interval T i . The spatio-temporal cube can reflect the continuous homogeneous distribution state of ground objects in a specific spatio-temporal domain of time-series images, thus ensuring the consistency of the distribution of geographical objects in space and time and making it easier to model and express the spatio-temporal correlation characteristics between geographical objects. The correlation distribution between geographical objects can be quantitatively expressed through the spatio-temporal neighborhood analysis of the spatio-temporal cube. As Figure 11 shown, the spatio-temporal cube C 1 reflects the homogeneous crop growth state of a farmland coverage area with a spatial range of S 1 within the time interval ΔT 1 (from September 12, 2017 to September 22, 2017). Similarly, C 2 reflects the homogeneous crop growth state of a farmland coverage area with a spatial range of S 2 within the time interval ΔT 2 , while C 3 reflects that the farmland coverage area with a spatial range of S 3 is in the state of crops being harvested within the time interval ΔT 2 . In particular, C 2 and C 3 are adjacent in space and both are adjacent to C 1 in time. That is to say, within the adjacent time periods ΔT 2 , C 1 is divided into two spatio-temporally adjacent spatio-temporal cubes C 2 and C 3 , reflecting the spatio-temporal change situation of the farmland in the S 1 area within ΔT 1 to ΔT 2 , that is, part of the S 3 area is harvested, while the S 2 area remains unchanged, reflecting the real change situation of the surface of this area. Therefore, the spatio-temporal context correlation between ground objects can be modeled by studying the adjacent correlation between spatio-temporal cubes, which helps the spatio-temporal correlation and change analysis of ground objects in time-series images. Specifically, from the characteristics of time-series images, it can be known that time-series image segmentation is oriented to a three-dimensional space composed of spatial and temporal dimensions. For each currently processed spatio-temporal cube, its adjacent cubes include both spatial neighborhoods and temporal neighborhoods. Therefore, asFigure 12 As shown, when the spatio-temporal cube is merged and expanded, it is necessary to consider the merger between adjacent cubes in space and also the merger and expansion between adjacent cubes in the time dimension. Therefore, it is necessary to comprehensively consider the heterogeneity differences before and after the merger in both the time and space domains. Therefore, in the present invention, the heterogeneity criterion for temporal image segmentation consists of two parts: spatial heterogeneity S h and temporal heterogeneity T h , and is formally expressed as follows:

[0173] H h ={S h , T h} (22)

[0174] Among them, T h refers to the heterogeneity difference between the current spatio-temporal cube and its temporal neighborhood, and S h is the heterogeneity difference between the current spatio-temporal cube and its spatial neighborhood. The spatio-temporal segmentation heterogeneity criterion is not a simple criterion for "merging" or "not merging" a pair of spatio-temporal solid objects, but is based on calculating the magnitude of the heterogeneity difference before and after the merger between the calculation objects, and this value can represent the matching degree of two spatio-temporal cubes.

[0175] In one embodiment, as Figure 13 shown, the heterogeneity difference in the spatial domain is obtained through the spectral heterogeneity difference and shape heterogeneity in space. When calculating the spectral heterogeneity difference in space for a pair of spatio-temporal cubes, it is necessary to calculate the spectral differences at all time points included in the spatio-temporal cube Then, the spectral standard deviation of each time point is obtained, and the average value is taken over the number of time points of the spatio-temporal solid to form the average spectral standard deviation h k of each spatio-temporal cube. At the same time, in order to ensure the relative consistency of the spatial heterogeneity calculation of different spatio-temporal cubes, the spectral heterogeneity h k of each cube is weighted using the volume size of the spatio-temporal cube. The volume is obtained by multiplying the area size of the spatial domain of the spatio-temporal cube, that is, the number of pixels in the spatial domain polygon, and the time interval length ΔT. Similarly, for the differences in compactness Cpt and smoothness Smooth in shape heterogeneity, the average values at all time points included in the spatio-temporal cube also need to be calculated. Therefore, it is necessary to sum the compactness and smoothness of each time layer and then take the average over the time interval length (ΔT) of the spatio-temporal cube. Similarly, in order to ensure the relative consistency of the shape heterogeneity calculation of different cubes, the differences in compactness and smoothness of each spatio-temporal cube are weighted using its volume size.

[0176] In one embodiment, a spatio-temporal region growing algorithm is used to process temporal images. Specifically, for the spatio-temporal cube composed of a single pixel in the temporal image, repetitive merging processing is performed on the single-pixel cube and its neighboring pixels until the spatio-temporal heterogeneity between the cubes reaches a given maximum heterogeneity parameter. This maximum heterogeneity parameter can be referred to as the spatio-temporal segmentation scale because it represents the maximum tolerance for the spatio-temporal heterogeneity of the spatio-temporal cube. By setting different spatio-temporal segmentation scale parameters, spatio-temporal cube results of different sizes can be obtained. Based on the spatio-temporal heterogeneity criterion, the time and space scale parameters can be defined respectively. As Figure 14 shown, in the spatio-temporal region growing algorithm, the spatial neighborhood and the time neighborhood of the spatio-temporal cube are determined according to the spatio-temporal neighborhood rule, and the optimal matching cube in the neighborhood of the current spatio-temporal cube is determined according to the heuristic merging rule of the optimal matching strategy. If the heterogeneity difference between the optimal matching neighborhood cube and the current spatio-temporal cube does not meet the given maximum heterogeneity parameter, the cube growth ends, and the neighborhood cube is used as the new processing object for region growth processing. If the heterogeneity difference between the optimal matching neighborhood cube and the current spatio-temporal cube meets the given maximum heterogeneity parameter, a merging operation is performed and the spatio-temporal cube grows. As Figure 15 shown, after growth, the spatio-temporal cube returns to the step of finding the optimal matching cube in the neighborhood of the current spatio-temporal cube according to the spatio-temporal neighborhood rule and the heuristic merging rule of the optimal matching strategy in the spatio-temporal region growing algorithm, and continues the spatio-temporal growth processing. In each algorithm cycle, each spatio-temporal cube will perform a growth judgment process. When the growth heterogeneity of all spatio-temporal cubes does not meet the maximum heterogeneity parameter, the algorithm ends and outputs the spatio-temporal cube segmentation result.

[0177] In one embodiment, equally spaced spatio-temporal scale parameters are used to generate spatio-temporal cubes, where smaller spatio-temporal scales are used to generate smaller cubes; while larger scales generate cubes with larger spatio-temporal domains. As Figure 16 shown, the large-scale spatio-temporal cube is composed of multiple small-scale cubes. Therefore, there is a strict one-to-many relationship between adjacent scales of spatio-temporal cubes. In this case, the spatio-temporal boundary of the large-scale spatio-temporal cube should be consistent with the spatio-temporal boundary of the small-scale cube, and the volume of each large-scale spatio-temporal cube is the sum of the volumes of several small-scale cubes. To achieve hierarchical multi-scale segmentation, the smallest-scale spatio-temporal cubes are directly generated by segmenting the pixels of the temporal image, while other larger-scale spatio-temporal cubes are generated by aggregating small-scale cubes. The hierarchical structure is more efficient for generating multi-scale spatio-temporal cube segmentation results, and the scale hierarchical relationship is also important for analyzing spatio-temporal cubes.

[0178] In one embodiment, during the segmentation of time-series images, after the spatio-temporal heterogeneity criterion and the multi-scale organization method are given, two spatio-temporal cubes to be merged are determined. For the merging and expansion process of each spatio-temporal cube, the spatial neighborhood of the cube is considered while also considering the temporal neighborhood of the cube. Whether to merge and expand the cube or create a new spatio-temporal cube is determined by the temporal and spatial neighborhoods of the spatio-temporal cube. As Figure 17 shown, for a spatio-temporal cube C 0 (S 0 ,T 0 ), its spatial neighborhood is a set of spatio-temporal cube collections {C i (S i ,T i )}, and the cubes in the collection should meet the following conditions:

[0179] (1) C i and C 0 are directly adjacent in the spatial domain, that is, C i and C 0 have a common face;

[0180] (2) The temporal domain of C i intersects with the temporal domain of C 0 , that is

[0181] Specifically, as Figure 18 shown, the temporal neighborhood of the spatio-temporal cube adopts the information of the nearest temporal points before and after. For the spatio-temporal cube C 0 (S 0 ,T 0 ), its temporal neighborhood is the cube collection {C i (S i ,T i )}, and meets the conditions:

[0182] (1) C i and C 0 are directly adjacent in time, that is, C i is in the nearest previous or next temporal phase of C 0 (T i = T 0 - 1 or T i = T 0 + 1);

[0183] (2) The spatial domain of C i and the spatial domain of C 0 have an intersection, that is

[0184] For any spatio-temporal cube C 0 (S 0 ,T 0) Multiple adjacent cubes C can be obtained in terms of both space and time through the spatio-temporal neighborhood rule i (S i ,T i ). Each neighborhood cube C i will calculate the spatio-temporal heterogeneity difference values before and after merging with C 0 , and all of them may be merged with C 0 . Therefore, it is necessary to select the optimal C i from the adjacent spatio-temporal cubes for merging, so as to ensure that the spatio-temporal heterogeneity difference brought by the merging of C i and C 0 is minimized. The decision-making basis for selection is the spatio-temporal heterogeneity difference values of C 0 with respect to each adjacent cube. Different decision-making methods will lead to different neighborhood cubes being merged with C 0 . The current spatio-temporal cube C 0 will be merged with the cube with the optimal heterogeneity in the spatio-temporal neighborhood. The optimal heterogeneity means that the heterogeneity difference between the neighborhood spatio-temporal cube and C 0 is less than the segmentation scale, and compared with the spatio-temporal heterogeneity differences of other neighborhood cubes, the spatio-temporal heterogeneity difference of the neighborhood cube to be merged is the smallest.

[0185] In one embodiment, as Figure 19 shown, experiments are carried out using time-series image data composed of multiple temporally consecutive remote sensing images. These images are roughly evenly distributed within one year from April 2014 to May 2015. The distribution information of the image time points is shown in Table 1, and the spatial resolution is 15m. The area is the suburban area of City A, and the image size is 1156×1096. This area covers typical land cover types such as cultivated land, buildings, forests, grasslands, water bodies, and bare soil, which is suitable for verifying the effectiveness of the method proposed in the present invention. The time-series segmentation method provided by the present invention is used for segmentation. In order to obtain segmentation results at multiple spatio-temporal scales, 5 spatio-temporal scales with gradually increasing gradients are selected for segmentation experiments, and the corresponding spatio-temporal scale parameters are shown in Table 2.

[0186] Table 1 Time-series image data

[0187]

[0188] In the above Table 1, the numbers in parentheses represent the number of days the data was obtained in a year. For example, 14119 means the image was obtained on the 119th day of 2014.

[0189] Table 2 Time-series image parameter table

[0190]

[0191] Taking the segmentation results at scale 3 as an example, a total of 34,089 spatio-temporal cubes were generated. These cubes are continuously distributed in the three-dimensional spatio-temporal reference system but do not intersect with each other. Each spatio-temporal cube, as a set of homogeneous pixels, is relatively uniform in a certain continuous spatio-temporal domain. As Figure 19 shown, the details of the segmentation results of some spatio-temporal cubes are presented. The length of the △T time domain is 32 days, that is, from time point 14119 to 14151, which reflects the homogeneous continuous distribution state of the corresponding ground objects of this spatio-temporal cube within this spatio-temporal range. In particular, as Figure 20 shown, two sub-regions A and B covered by farmland are selected, and the details of the segmentation results of the spatio-temporal cubes from time point 14279 to time point 15106 are presented. For sub-region A, 3 spatio-temporal cubes A 1 、A 2 and A 3 are generated after segmentation. They have the same spatial domain distribution but different time domain information. At the same time, for sub-region B, 4 spatio-temporal cubes B 1 、B 2 、B 3 and B 4 are generated, and there are changes in both time and spatial domains among them. The changes occurring in region B are caused by human agricultural planting activities. Some farmlands will be covered with plastic greenhouses in winter to ensure the overwintering of crops. Therefore, the farmlands in regions A and B change from the crop growth state, that is, A 1 、B 1 and B 2 are transformed into the greenhouse-covered state, that is, A 2 and B 3 . The results show that the proposed spatio-temporal remote sensing image segmentation method has obtained good results at scale 3, and the spatio-temporal distribution and change information of geographical objects can be well reflected by the segmented spatio-temporal cubes. This spatio-temporal remote sensing image segmentation method can be encapsulated into a spatio-temporal cube model and applied as a basic unit to the classification and change analysis of ground objects in spatio-temporal remote sensing images. The segmentation results of spatio-temporal cubes can serve as the basis for the classification and change analysis of ground objects in spatio-temporal remote sensing images.

[0192] The spatio-temporal remote sensing image segmentation device provided by the present invention will be described below. The spatio-temporal remote sensing image segmentation device described below can be mutually referred to with the spatio-temporal remote sensing image segmentation method described above.

[0193] In one embodiment, the present invention further provides a time-series remote sensing image segmentation device 2100, including a target time-series remote sensing image acquisition module 2102, a multi-dimensional feature sequence obtaining module 2104, a target segmentation spatio-temporal cube obtaining module 2106, and a segmented time-series remote sensing image obtaining module 2108, where: The target time-series remote sensing image acquisition module 2102 is configured to acquire a target time-series remote sensing image to be segmented; The multi-dimensional feature sequence obtaining module 2104 is configured to perform multi-dimensional feature analysis on each pixel point in the target time-series remote sensing image to obtain a multi-dimensional feature sequence; The target segmentation spatio-temporal cube obtaining module 2106 is configured to obtain a target segmentation spatio-temporal cube for each pixel point in the target time-series remote sensing image based on the multi-dimensional feature matching degree between the multi-dimensional features in the multi-dimensional feature sequence; The segmented time-series remote sensing image obtaining module 2108 is configured to segment the target time-series remote sensing image by using the target segmentation spatio-temporal cube to obtain a time-series remote sensing image segmentation result.

[0194] In one embodiment, the target segmentation spatio-temporal cube obtaining module 2106 is configured to obtain an intermediate spatio-temporal cube for each pixel point in the target time-series remote sensing image based on the multi-dimensional feature matching degree between the multi-dimensional features in the multi-dimensional feature sequence; In the intermediate spatio-temporal cube, a candidate spatio-temporal cube is determined; A target segmentation spatio-temporal cube is obtained for the candidate pixel points based on the multi-dimensional feature matching degree between the candidate spatio-temporal cube and a comparison spatio-temporal cube; where the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube.

[0195] In one embodiment, the target segmentation spatio-temporal cube obtaining module 2106 is configured to obtain a first spectral feature matching degree and a shape feature matching degree in the spatial dimension matching degree; perform a weighted sum of the first spectral feature matching degree and the shape feature matching degree to obtain a first target matching degree; obtain a target segmentation spatio-temporal cube for each pixel point in the target time-series remote sensing image based on the first target matching degree.

[0196] In one embodiment, the target segmentation spatio-temporal cube obtaining module 2106 is configured to obtain a pixel point sequence corresponding to the target time-series remote sensing image; the pixel point sequence is a sequence composed of each pixel point in the target time-series remote sensing image; in the pixel point sequence, a candidate spatio-temporal cube is determined; when the first target matching degree is greater than a preset matching degree threshold, the candidate spatio-temporal cube and a comparison spatio-temporal cube are merged to obtain a merged spatio-temporal cube; the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube; the merged spatio-temporal cube is determined as the candidate spatio-temporal cube, and the step of when the first target matching degree is greater than a preset matching degree threshold, the candidate spatio-temporal cube and a comparison spatio-temporal cube are merged to obtain a merged spatio-temporal cube is executed.

[0197] In one embodiment, the target segmentation spatio-temporal cube obtaining module 2106 is configured to obtain a first compactness value and a first smoothness value corresponding to a candidate spatio-temporal cube, and obtain a second compactness value and a second smoothness value corresponding to a comparison spatio-temporal cube; and obtain a third compactness value and a third smoothness value of the merged spatio-temporal cube; based on the first compactness value, the first smoothness value, the second compactness value, the second smoothness value, the third compactness value and the third smoothness value, obtain a compactness difference value and a smoothness difference value; perform a weighted sum on the compactness difference value and the smoothness difference value to obtain a shape difference value; based on the correspondence between the shape difference value and the shape feature matching degree, obtain the shape feature matching degree.

[0198] In one embodiment, the target segmentation spatio-temporal cube obtaining module 2106 is configured to obtain the number of spectral bands in the target temporal remote sensing image; based on the number of spectral bands, obtain a first spectral feature matching degree.

[0199] In one embodiment, the target segmentation spatio-temporal cube obtaining module 2106 is configured to obtain a second spectral feature matching degree in the time dimension matching degree; based on the second spectral feature matching degree, for each pixel point in the target temporal remote sensing image, obtain the target segmentation spatio-temporal cube.

[0200] Figure 22 The schematic physical structure diagram of an electronic device is exemplified, as Figure 22 shown, the electronic device may include: a processor 2210, a communication interface 2220, a memory 2230, and a communication bus 2240. Among them, the processor 2210, the communication interface 2220, and the memory 2230 complete mutual communication through the communication bus 2240. The processor 2210 may call logic instructions in the memory 2230 to execute a temporal remote sensing image segmentation method, and the method includes: obtaining a target temporal remote sensing image to be segmented; performing multi-dimensional feature analysis on each pixel point in the target temporal remote sensing image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between the multi-dimensional features in the multi-dimensional feature sequence, for each pixel point in the target temporal remote sensing image, obtain a target segmentation spatio-temporal cube; using the target segmentation spatio-temporal cube to segment the target temporal remote sensing image to obtain a segmented temporal remote sensing image.

[0201] In addition, when the logical instructions in the above-mentioned memory 2230 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0202] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the time-series remote sensing image segmentation method provided by the above-mentioned various methods. The method includes: acquiring a target time-series remote sensing image to be segmented; performing multi-dimensional feature analysis on each pixel point in the target time-series remote sensing image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, for each pixel point in the target time-series remote sensing image, obtaining a target segmentation spatio-temporal cube; using the target segmentation spatio-temporal cube to segment the target time-series remote sensing image to obtain a segmented time-series remote sensing image.

[0203] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the time-series remote sensing image segmentation method provided by the above-mentioned various methods. The method includes: acquiring a target time-series remote sensing image to be segmented; performing multi-dimensional feature analysis on each pixel point in the target time-series remote sensing image to obtain a multi-dimensional feature sequence; based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, for each pixel point in the target time-series remote sensing image, obtaining a target segmentation spatio-temporal cube; using the target segmentation spatio-temporal cube to segment the target time-series remote sensing image to obtain a segmented time-series remote sensing image.

[0204] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative work.

[0205] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0206] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for segmenting temporal remote sensing images, characterized in that, it includes: Obtain the target temporal remote sensing image to be segmented; Perform multi-dimensional feature analysis on each pixel point in the target temporal remote sensing image to obtain a multi-dimensional feature sequence; Based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, process each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube sequence, including: Based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, process each pixel point in the target temporal remote sensing image to obtain an intermediate spatio-temporal cube sequence; In the intermediate spatio-temporal cube sequence, determine candidate spatio-temporal cubes, where the candidate spatio-temporal cubes are either a pixel point arbitrarily selected from each pixel point of the target temporal remote sensing image or a pixel point set with a candidate identifier; Based on the multi-dimensional feature matching degree between the candidate spatio-temporal cube and the comparison spatio-temporal cube, process the candidate spatio-temporal cube to obtain a target pixel segmentation spatio-temporal cube; where the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube; Use the target segmentation spatio-temporal cube sequence to segment the target temporal remote sensing image to obtain a temporal remote sensing image segmentation result; Among them, the multi-dimensional feature matching degree includes a spatial dimension matching degree, and based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, processing each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube sequence includes: Obtain the first spectral feature matching degree and shape feature matching degree in the spatial dimension matching degree; Perform weighted summation on the first spectral feature matching degree and the shape feature matching degree to obtain a first target matching degree; Based on the first target matching degree, process each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube sequence.

2. The method for segmenting temporal remote sensing images according to claim 1, characterized in that, The processing each pixel point in the target temporal remote sensing image based on the first target matching degree to obtain a target segmentation spatio-temporal cube sequence includes: Obtain the pixel point sequence corresponding to the target temporal remote sensing image; the pixel point sequence is a sequence composed of each pixel point in the target temporal remote sensing image; In the pixel point sequence, determine candidate spatio-temporal cubes; If the first target matching degree is greater than a preset matching degree threshold, merge the candidate spatio-temporal cube with a comparison spatio-temporal cube to obtain a merged spatio-temporal cube, forming an initial spatio-temporal cube; the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube; Determine the merged spatio-temporal cube as the candidate spatio-temporal cube, and execute the step of if the first target matching degree is greater than a preset matching degree threshold, merge the candidate spatio-temporal cube with a comparison spatio-temporal cube to obtain a merged spatio-temporal cube.

3. The method for segmenting temporal remote sensing images according to claim 2, characterized in that, The process of obtaining the shape feature matching degree includes: Obtaining a first compactness value and a first smoothness value corresponding to the candidate spatio-temporal cube, and obtaining a second compactness value and a second smoothness value corresponding to the comparison spatio-temporal cube; and obtaining a third compactness value and a third smoothness value of the merged spatio-temporal cube; Based on the first compactness value, the first smoothness value, the second compactness value, the second smoothness value, the third compactness value and the third smoothness value, obtaining a compactness difference value and a smoothness difference value; Performing weighted summation on the compactness difference value and the smoothness difference value to obtain a shape difference value; Based on the corresponding relationship between the shape difference value and the shape feature matching degree, obtaining the shape feature matching degree.

4. The method for segmenting time-series remote sensing images according to claim 1, wherein, the process of obtaining the first spectral feature matching degree includes: Obtaining the number of spectral bands in the target time-series remote sensing image; Based on the number of spectral bands, obtaining the first spectral feature matching degree.

5. The method for segmenting time-series remote sensing images according to claim 1, wherein, the multi-dimensional feature matching degree includes a time dimension matching degree, and based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence, processing each pixel point in the target time-series remote sensing image to obtain a target segmentation spatio-temporal cube sequence, including: Obtaining a second spectral feature matching degree in the time dimension matching degree; Based on the second spectral feature matching degree, for each pixel point in the target time-series remote sensing image, obtaining a target pixel segmentation spatio-temporal cube.

6. A device for segmenting time-series remote sensing images, wherein, it includes: A target time-series remote sensing image acquisition module, configured to acquire a target time-series remote sensing image to be segmented; A multi-dimensional feature sequence obtaining module, configured to perform multi-dimensional feature analysis on each pixel point in the target time-series remote sensing image to obtain a multi-dimensional feature sequence; A target pixel segmentation spatio-temporal cube obtaining module, configured to process each pixel point in the target time-series remote sensing image based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence to obtain a target segmentation spatio-temporal cube sequence, including: Processing each pixel point in the target time-series remote sensing image based on the multi-dimensional feature matching degree between each multi-dimensional feature in the multi-dimensional feature sequence to obtain an intermediate spatio-temporal cube sequence; In the intermediate spatio-temporal cube sequence, determining a candidate spatio-temporal cube, where the candidate spatio-temporal cube is a pixel point randomly selected from each pixel point of the target time-series remote sensing image or a pixel point set with a candidate identifier; Processing the candidate spatio-temporal cube based on the multi-dimensional feature matching degree between the candidate spatio-temporal cube and the comparison spatio-temporal cube to obtain a target pixel segmentation spatio-temporal cube; where the comparison spatio-temporal cube is a spatio-temporal cube within the neighborhood of the candidate spatio-temporal cube; A module for segmenting temporal remote sensing images, which is used to segment the target temporal remote sensing image by using the target segmentation spatio-temporal cube sequence to obtain a temporal remote sensing image segmentation result; Wherein, the multi-dimensional feature matching degree includes a spatial dimension matching degree, and processing each pixel point in the target temporal remote sensing image based on the multi-dimensional feature matching degrees between the multi-dimensional features in the multi-dimensional feature sequence to obtain the target segmentation spatio-temporal cube sequence includes: Obtain the first spectral feature matching degree and the shape feature matching degree in the spatial dimension matching degree; Perform weighted summation on the first spectral feature matching degree and the shape feature matching degree to obtain a first target matching degree; Based on the first target matching degree, process each pixel point in the target temporal remote sensing image to obtain a target segmentation spatio-temporal cube sequence.

7. An electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, Characterized in that, When the processor executes the program, the steps of the temporal remote sensing image segmentation method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium, on which a computer program is stored, Characterized in that, When the computer program is executed by a processor, the steps of the temporal remote sensing image segmentation method according to any one of claims 1 to 5 are implemented.

Images