Remote sensing image semantic segmentation model migration learning method and device and computer equipment

Abstract

Embodiments of the present application disclose a remote sensing image semantic segmentation model transfer learning method and device and a computer device. The method comprises: performing initial training on a remote sensing image semantic segmentation model using a public fine annotation dataset to obtain a first model; performing data enhancement on the fine annotation data according to a target dataset to obtain an enhanced dataset; training the first model using the enhanced dataset to obtain a second model; annotating the target dataset, and fine-tuning the second model using the annotated target dataset. The method of the embodiments of the present application can process the existing public fine annotation dataset, so that the remote sensing image semantic segmentation model can obtain higher precision under the condition that the annotation amount of the target dataset is small.

Description

TECHNICAL FIELD

[0001] The present application relates to a semantic segmentation model, more particularly to a remote sensing image semantic segmentation model transfer learning method, device and computer equipment. BACKGROUND

[0002] Remote sensing image refers to a film or photo recording the electromagnetic wave size of various ground objects, mainly divided into aerial photos and satellite photos.

[0003] Due to the diversity of ground object types and the high cost of sample labeling, in order to quickly and widely perform semantic segmentation on remote sensing images, a transfer learning method is often used in practice. In the past, there have been a large number of studies on transfer learning methods. The basic process of the transfer learning method for the remote sensing image semantic segmentation model is as follows: a model is trained using a data set similar to the target data set; the model is modified as needed, such as adding or deleting relevant processing layers; part of the samples on the target data set are labeled; and the trained model is fine-tuned using the labeled part of the samples on the target data set.

[0004] Due to the small number of publicly available fine-labeled remote sensing data sets, there is often a large difference between the target data set and the fine-labeled data set, usually including resolution, deformation and color tone, so a large amount of labeling of the target data set is often required to make the model meet the current data situation.

[0005] Therefore, it is necessary to design a new method to achieve higher precision of the remote sensing image semantic segmentation model under the condition of less labeling of the target data set. SUMMARY

[0006] The present application aims to overcome the defects of the prior art and provide a remote sensing image semantic segmentation model transfer learning method, device and computer equipment.

[0007] To achieve the above-mentioned purpose, the present application adopts the following technical solution: a remote sensing image semantic segmentation model transfer learning method, comprising:

[0008] using a publicly available fine-labeled data set to perform initial training of a remote sensing image semantic segmentation model to obtain a first model;

[0009] performing data augmentation on the fine-labeled data according to the target data set to obtain an augmented data set;

[0010] training the first model using the augmented data set to obtain a second model;

[0011] labeling the target data set and fine-tuning the second model using the labeled target data set.

[0012] Further technical solutions thereof are that the resolution adjustment is performed on the fine-labeled data according to a target data set to obtain resolution adjustment data, and the resolution adjustment comprises:

[0013] The resolution adjustment is performed on the fine-labeled data according to a target data set to obtain resolution adjustment data.

[0014] The morphing is performed on the resolution adjustment data to obtain morphing data.

[0015] The tone adjustment is performed on the morphing data to obtain the enhanced data set.

[0016] Further technical solutions thereof are that the resolution adjustment is performed on the fine-labeled data according to a target data set to obtain resolution adjustment data, and the resolution adjustment comprises:

[0017] The fine-labeled data set samples are resampled and cropped according to the resolution of the target data set to obtain the resolution adjustment data.

[0018] Further technical solutions thereof are that the morphing is performed on the resolution adjustment data to obtain morphing data, and the morphing comprises:

[0019] A polynomial model method is adopted to construct a relationship between real space positions and image space positions for the target data set.

[0020] The relationship between the real space positions and the image space positions is applied to the resolution adjustment data to obtain the morphing data.

[0021] Further technical solutions thereof are that the tone adjustment is performed on the morphing data to obtain the enhanced data set, and the tone adjustment comprises:

[0022] A histogram matching method is adopted to adjust a histogram of the morphing data to be similar to a histogram of the target data set to obtain the enhanced data set.

[0023] The application further provides a remote sensing image semantic segmentation model transfer learning device, comprising:

[0024] A primary training unit is configured to perform primary training on a remote sensing image semantic segmentation model using a public fine-labeled data set to obtain a first model.

[0025] A data enhancement unit is configured to perform data enhancement on the fine-labeled data according to a target data set to obtain an enhanced data set.

[0026] A secondary training unit is configured to perform training on the first model using the enhanced data set to obtain a second model.

[0027] A fine-tuning unit is configured to label the target data set and fine-tune the second model by using the labeled target data set.

[0028] Further, the data augmentation unit comprises:

[0029] A resolution adjustment subunit is configured to adjust the resolution of the fine-labeled data according to the target data set to obtain resolution adjustment data.

[0030] A deformation subunit is configured to deform the resolution adjustment data to obtain deformation data.

[0031] A tone adjustment subunit is configured to adjust the tone of the deformation data to obtain an augmented data set.

[0032] Further, the resolution adjustment subunit is configured to resample and crop the fine-labeled data set sample according to the resolution of the target data set to obtain the resolution adjustment data.

[0033] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0034] The application further provides a storage medium storing a computer program, wherein the computer program is executed by a processor to implement the above method.

[0035] Compared with the prior art, the application has the following beneficial effects: the model is initially trained by using the disclosed fine-labeled data set, the fine-labeled data is augmented according to the target data set, the first model after initial training is trained by using the augmented data set, the target data set is labeled, the second model is fine-tuned by using the labeled target data set, the disclosed fine-labeled data set is processed, and the remote sensing image semantic segmentation model has higher precision under the condition that the target data set has less label quantity.

[0036] The application will be further described below in combination with the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative effort.

[0038] Figure 1An application scenario schematic diagram of the remote sensing image semantic segmentation model transfer learning method provided by the embodiment of the present application is provided.

[0039] Figure 2 A flowchart of the remote sensing image semantic segmentation model transfer learning method provided by the embodiment of the present application is provided.

[0040] Figure 3 A sub-flowchart of the remote sensing image semantic segmentation model transfer learning method provided by the embodiment of the present application is provided.

[0041] Figure 4 A sub-flowchart of the remote sensing image semantic segmentation model transfer learning method provided by the embodiment of the present application is provided.

[0042] Figure 5 A schematic block diagram of the remote sensing image semantic segmentation model transfer learning device provided by the embodiment of the present application is provided.

[0043] Figure 6 A schematic block diagram of the data enhancement unit of the remote sensing image semantic segmentation model transfer learning device provided by the embodiment of the present application is provided.

[0044] Figure 7 A schematic block diagram of the deformation sub-unit of the remote sensing image semantic segmentation model transfer learning device provided by the embodiment of the present application is provided.

[0045] Figure 8 A schematic block diagram of the computer device provided by the embodiment of the present application is provided. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0047] It should be understood that when used in the specification and the appended claims, the terms “comprise” and “include” indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0048] It should also be understood that the terms used herein in the specification and the appended claims are only for the purpose of describing particular embodiments and do not intend to limit the present application. As used in the specification and the appended claims of the present application, unless otherwise clearly indicated by the context, the singular forms “a”, “an” and “the” are intended to include the plural forms as well.

[0049] It should be further understood that the term "and / or" used in the description and claims of the application means one or more of the associated listed items as well as all possible combinations of the items and includes these combinations.

[0050] Referring to Figure 1 and Figure 2 , Figure 1 The application scenario schematic diagram of the remote sensing image semantic segmentation model transfer learning method provided by the embodiment of the application. Figure 2 The schematic flowchart of the remote sensing image semantic segmentation model transfer learning method provided by the embodiment of the application. The remote sensing image semantic segmentation model transfer learning method is applied to a server. The server interacts with a terminal to perform data, trains a model by using a public fine annotation data set, performs data enhancement on the fine annotation data according to a target data set, trains a first model after initial training by using the enhanced data set, annotates the target data set, and fine-tunes the second model by using the annotated target data set, so as to process the existing public fine annotation data set, and make the model obtain higher precision in the case that the annotation amount of the target data set is small.

[0051] Figure 2 The flowchart of the remote sensing image semantic segmentation model transfer learning method provided by the embodiment of the application. As shown in Figure 2 The method comprises the following steps S110 to S140.

[0052] S110, a remote sensing image semantic segmentation model is initially trained by using a public fine annotation data set to obtain a first model.

[0053] In the embodiment, the first model refers to a model formed after the remote sensing image semantic segmentation model is initially trained by using the public fine annotation data set.

[0054] S120, the fine annotation data is enhanced according to the target data set to obtain an enhanced data set.

[0055] In the embodiment, the enhanced data set refers to a data set formed after the fine annotation data is enhanced according to the target data set.

[0056] In an embodiment, referring to Figure 3 The step S120 can comprise steps S121 to S123.

[0057] S121, the resolution of the fine annotation data is adjusted according to the target data set to obtain resolution adjustment data.

[0058] In the embodiment, the resolution adjustment data refers to a data set formed after resampling and cropping the fine annotation data set sample according to the resolution of the target data set.

[0059] Specifically, the fine annotation data set sample is resampled and cropped according to the resolution of the target data set to obtain the resolution adjustment data.

[0060] S122, morphing the resolution adjustment data to obtain morphed data.

[0061] In the embodiment, the morphed data refers to data formed after morphing the resolution adjustment data.

[0062] In an embodiment, referring to Figure 4 The step S122 can include steps S1221-S1222.

[0063] S1221, using a polynomial model method, constructing a relationship between real space positions and image space positions for the target data set;

[0064] S1222, applying the relationship between real space positions and image space positions to the resolution adjustment data to obtain morphed data.

[0065] Using a polynomial model method, constructing a relationship between real space positions (x_true, y_true) and image space positions (x_obs, y_obs) for the target data set: (x_obs, y_obs) = f(x_true, y_true: a_ij, b_ij); such as a polynomial correction model, and applying f(x_true, y_true) to the public fine annotation data set to generate the same morphed data.

[0066] Specifically, the polynomial correction model is: Where (x, y) and (u, v) are the position information of the same named points in the two images, a ij and b ij are polynomial parameters, and the optimal parameters are solved by annotation control points.

[0067] S123, adjusting the tone of the morphed data to obtain an enhanced data set.

[0068] In the embodiment, a histogram matching method is used to adjust the histogram of the morphed data to be similar to that of the target data set to obtain an enhanced data set.

[0069] S130, training the first model using the enhanced data set to obtain a second model;

[0070] S140, label the target data set, and fine-tune the second model using the labeled target data set.

[0071] In the embodiment, the training and fine-tuning of the model belong to the prior art, which will not be described here.

[0072] The remote sensing image semantic segmentation model transfer learning method described above uses a public fine-labeled data set to initially train a model, and performs data enhancement on the fine-labeled data according to the situation of the target data set, uses the enhanced data set to train the first model after initial training, labels the target data set, and fine-tunes the second model using the labeled target data set, realizes processing of the existing public fine-labeled data set, and makes the remote sensing image semantic segmentation model obtain higher precision in the case of less target data set labeling.

[0073] Figure 5 is a schematic block diagram of a remote sensing image semantic segmentation model transfer learning device 300 provided by an embodiment of the application. As Figure 5 shown, corresponding to the above remote sensing image semantic segmentation model transfer learning method, the application further provides a remote sensing image semantic segmentation model transfer learning device 300. The remote sensing image semantic segmentation model transfer learning device 300 includes units for executing the above remote sensing image semantic segmentation model transfer learning method, and the device can be configured in a server. Specifically, please refer to Figure 5 , the remote sensing image semantic segmentation model transfer learning device 300 includes an initial training unit 301, a data enhancement unit 302, a secondary training unit 303, and a fine-tuning unit 304.

[0074] The initial training unit 301 is configured to use a public fine-labeled data set to initially train a remote sensing image semantic segmentation model to obtain a first model; the data enhancement unit 302 is configured to perform data enhancement on the fine-labeled data according to the situation of the target data set to obtain an enhanced data set; the secondary training unit 303 is configured to use the enhanced data set to train the first model to obtain a second model; and the fine-tuning unit 304 is configured to label the target data set and fine-tune the second model using the labeled target data set.

[0075] In an embodiment, as Figure 6 shown, the data enhancement unit 302 includes a resolution adjustment subunit 3021, a deformation subunit 3022, and a color tone adjustment subunit 3023.

[0076] The resolution adjustment sub-unit 3021 is configured to adjust the resolution of the fine annotation data according to the target data set to obtain resolution adjustment data; the morphing sub-unit 3022 is configured to morph the resolution adjustment data to obtain morphed data; and the tone adjustment sub-unit 3023 is configured to adjust the tone of the morphed data to obtain an enhanced data set.

[0077] In an embodiment, the resolution adjustment sub-unit 3021 is configured to resample and crop the fine annotation data set sample according to the resolution of the target data set to obtain the resolution adjustment data.

[0078] In an embodiment, as shown in Figure 7 , the morphing sub-unit 3022 includes a relationship construction module 30221 and an application module 30222.

[0079] The relationship construction module 30221 is configured to construct the relationship between the real space position and the image space position for the target data set by using a polynomial model method; and the application module 30222 is configured to apply the relationship between the real space position and the image space position to the resolution adjustment data to obtain the morphed data.

[0080] In an embodiment, the tone adjustment sub-unit 3023 is configured to adjust the histogram of the morphed data to be similar to the histogram of the target data set by using a histogram matching method to obtain the enhanced data set.

[0081] It should be noted that the specific implementation process of the remote sensing image semantic segmentation model transfer learning device 300 and each unit can be clearly understood by those skilled in the art, and can refer to the corresponding description in the foregoing method embodiments. For the convenience and brevity of description, it will not be repeated here.

[0082] The remote sensing image semantic segmentation model transfer learning device 300 can be implemented in the form of a computer program, which can run on a computer device as shown in Figure 8 .

[0083] Please refer to Figure 8 , Figure 8 is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 500 can be a server, wherein the server can be a stand-alone server or a server cluster composed of multiple servers.

[0084] Referring to Figure 8 , the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501, wherein the memory can include a non-volatile storage medium 503 and an internal memory 504.

[0085] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a remote sensing image semantic segmentation model transfer learning method.

[0086] The processor 502 is configured to provide computing and control capabilities to support the operation of the entire computer device 500.

[0087] The memory 504 provides an environment for the computer program 5032 in the non-volatile storage medium 503 to run. When the computer program 5032 is executed by the processor 502, the processor 502 can perform a remote sensing image semantic segmentation model transfer learning method.

[0088] The network interface 505 is configured to communicate with other devices via a network. Those skilled in the art can understand that the network interface 505 can be configured to communicate with other devices via a wired or wireless network, and the network interface 505 can include a transceiver, a modem, a network interface card, etc. Figure 8 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device 500 to which the scheme of the present application is applied. The specific computer device 500 can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0089] The processor 502 is configured to run the computer program 5032 stored in the memory to implement the following steps:

[0090] The disclosed fine annotation dataset is used to perform initial training of the remote sensing image semantic segmentation model to obtain a first model. The fine annotation data is data-augmented according to the target dataset to obtain an augmented dataset. The first model is trained using the augmented dataset to obtain a second model. The target dataset is annotated, and the second model is fine-tuned using the annotated target dataset.

[0091] In an embodiment, when implementing the step of data-augmenting the fine annotation data according to the target dataset to obtain an augmented dataset, the processor 502 specifically implements the following steps:

[0092] The resolution of the fine annotation data is adjusted according to the target dataset to obtain resolution-adjusted data. The resolution-adjusted data is morphed to obtain morphed data. The morphed data is adjusted in tone to obtain the augmented dataset.

[0093] In an embodiment, when implementing the step of adjusting the resolution of the fine annotation data according to the target dataset to obtain resolution-adjusted data, the processor 502 specifically implements the following steps:

[0094] resampling the fine annotation dataset sample according to a resolution of the target dataset and cropping to obtain resolution adjusted data.

[0095] In an embodiment, the processor 502, when implementing the step of morphing the resolution adjusted data to obtain morphed data, implements the following steps:

[0096] using a polynomial model method, constructing a relationship between real space positions and image space positions for the target dataset; and applying the relationship between real space positions and image space positions to the resolution adjusted data to obtain morphed data.

[0097] In an embodiment, the processor 502, when implementing the step of tone adjusting the morphed data to obtain an enhanced dataset, implements the following steps:

[0098] using a histogram matching method to adjust a histogram of the morphed data to be similar to a histogram of the target dataset to obtain an enhanced dataset.

[0099] It should be understood that, in the embodiments of the present application, the processor 502 can be a central processing unit (CPU), and the processor 502 can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0100] It can be understood by those skilled in the art that all or part of the processes in the method of the above embodiments can be completed by a computer program instructing related hardware. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiments of the method.

[0101] Therefore, the present application also provides a storage medium. The storage medium can be a computer readable storage medium. The storage medium stores a computer program, wherein the computer program is executed by a processor to make the processor perform the following steps:

[0102] The first model is obtained by using a disclosed fine annotation dataset to perform initial training on a remote sensing image semantic segmentation model; the fine annotation data is data-augmented according to a target dataset to obtain an augmented dataset; the first model is trained using the augmented dataset to obtain a second model; the target dataset is annotated, and the second model is fine-tuned using the annotated target dataset.

[0103] In an embodiment, when the processor executes the computer program to implement the step of data-augmenting the fine annotation data according to the target dataset to obtain an augmented dataset, the following steps are implemented:

[0104] The resolution of the fine annotation data is adjusted according to the target dataset to obtain resolution-adjusted data; the resolution-adjusted data is morphed to obtain morphed data; and the morphed data is tone-adjusted to obtain the augmented dataset.

[0105] In an embodiment, when the processor executes the computer program to implement the step of adjusting the resolution of the fine annotation data according to the target dataset to obtain resolution-adjusted data, the following steps are implemented:

[0106] The fine annotation dataset samples are resampled and cropped according to the resolution of the target dataset to obtain the resolution-adjusted data.

[0107] In an embodiment, when the processor executes the computer program to implement the step of morphing the resolution-adjusted data to obtain morphed data, the following steps are implemented:

[0108] A polynomial model method is used to construct a relationship between real space positions and image space positions for the target dataset; and the relationship between real space positions and image space positions is applied to the resolution-adjusted data to obtain the morphed data.

[0109] In an embodiment, when the processor executes the computer program to implement the step of tone-adjusting the morphed data to obtain an augmented dataset, the following steps are implemented:

[0110] A histogram matching method is used to adjust the histogram of the morphed data to be similar to the histogram of the target dataset to obtain the augmented dataset.

[0111] The storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, and various computer-readable storage media that can store program codes.

[0112] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0113] In several embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic. For example, the division of each unit is only a logical function division, and actual implementation can have another division manner. For example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed.

[0114] The steps in the method embodiments of the present application can be adjusted, combined and deleted in sequence according to actual needs. The units in the apparatus embodiments of the present application can be combined, divided and deleted according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0115] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a terminal or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application.

[0116] The above description is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for remote sensing image semantic segmentation model transfer learning, characterized in that, The method comprises the following steps: performing initial training on a remote sensing image semantic segmentation model using a public fine annotation dataset to obtain a first model; performing data enhancement on the fine annotation dataset according to a target dataset to obtain an enhanced dataset; training the first model using the enhanced dataset to obtain a second model; annotating the target dataset and fine-tuning the second model using the annotated target dataset; the data enhancement on the fine annotation dataset according to the target dataset to obtain the enhanced dataset comprises: adjusting the resolution of the fine annotation dataset according to the target dataset to obtain resolution adjustment data; deforming the resolution adjustment data to obtain deformation data; adjusting the color tone of the deformation data to obtain the enhanced dataset. 2.The remote sensing image semantic segmentation model transfer learning method according to claim 1, characterized in that, the resolution adjustment of the fine annotation dataset according to the target dataset to obtain the resolution adjustment data comprises: resampling and cropping the fine annotation dataset samples according to the resolution of the target dataset to obtain the resolution adjustment data. 3.The method of claim 1, wherein, the deformation of the resolution adjustment data to obtain the deformation data comprises: using a polynomial model method, constructing the relationship between the real space position and the image space position for the target dataset; applying the relationship between the real space position and the image space position to the resolution adjustment data to obtain the deformation data. 4.The method of claim 1, wherein, the color tone adjustment of the deformation data to obtain the enhanced dataset comprises: using a histogram matching method to adjust the histogram of the deformation data to be similar to the histogram of the target dataset to obtain the enhanced dataset.

5. The remote sensing image semantic segmentation model transfer learning device, characterized in that, The method comprises the following steps: an initial training unit is configured to perform initial training on a remote sensing image semantic segmentation model using a public fine annotation dataset to obtain a first model; a data enhancement unit is configured to perform data enhancement on the fine annotation dataset according to a target dataset to obtain an enhanced dataset; the enhanced dataset comprises: adjusting the resolution of the fine annotation dataset according to the target dataset to obtain resolution adjustment data; deforming the resolution adjustment data to obtain deformation data; adjusting the color tone of the deformation data to obtain the enhanced dataset; a secondary training unit is configured to train the first model using the enhanced dataset to obtain a second model; a fine-tuning unit is configured to annotate the target dataset and fine-tune the second model using the annotated target dataset; the data enhancement unit comprises: a resolution adjustment subunit is configured to adjust the resolution of the fine annotation dataset according to the target dataset to obtain resolution adjustment data; a deformation subunit is configured to deform the resolution adjustment data to obtain deformation data; a color tone adjustment subunit is configured to adjust the color tone of the deformation data to obtain the enhanced dataset. 6.The remote sensing image semantic segmentation model transfer learning apparatus of claim 5, wherein, the resolution adjustment subunit is configured to resample and crop the fine annotation dataset samples according to the resolution of the target dataset to obtain the resolution adjustment data.

7. A computer device, comprising: The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1-4 when executing the computer program.

8. A storage medium, characterized by The storage medium stores a computer program, and the processor implements the method according to any one of claims 1-4 when executing the computer program.

Images