WO2023000159A1 - Semi-supervised classification method, apparatus and device for high-resolution remote sensing image, and medium - Google Patents

Abstract

A semi-supervised classification method, apparatus and device for a high-resolution remote sensing image, and a medium. The method comprises: pre-processing a remote sensing image; making a ground object classification sample set according to the processed remote sensing image; constructing a remote sensing image semantic segmentation model based on a Unet++ network, and training the semantic segmentation model by means of the ground object classification sample set; constructing a threshold value segmentation model based on a near-infrared band; performing model fusion on the semantic segmentation model and the threshold value segmentation model, so as to acquire a classification model; and classifying, by using the classification model, a remote sensing image to be classified. In this way, a remote sensing image semantic segmentation model based on a Unet++ network and a threshold value segmentation model based on a near-infrared band are constructed, then, texture information of a remote sensing image and spectral information of the near-infrared band are fused by using a multi-model fusion method, and then, a high-resolution remote sensing image is classified, such that the classification accuracy can be improved.

Description

High-resolution remote sensing image semi-supervised classification method, device, equipment and medium technical field

The invention relates to the field of classification of remote sensing images, in particular to a method, device, equipment and medium for semi-supervised classification of high-resolution remote sensing images.

Background technique

As the most important part of remote sensing satellite products, high-resolution remote sensing images are widely used in agricultural production estimation, agricultural risk assessment, mineral investigation, land and resources investigation and other fields. In recent years, high-resolution remote sensing satellites have been launched rapidly, and image data sources have increased dramatically, providing abundant data resources for later applications. The processing of remote sensing data generally includes pre-processing and advanced analysis. In post-application, image classification is the basic research to understand the coverage of ground objects.

Generally speaking, the resolution of high-resolution remote sensing images is better than 1 meter (ie > 1 meter), which can clearly distinguish vegetation, water bodies, buildings and other ground objects. Using high-resolution remote sensing images to classify ground features can grasp the details of ground features and understand the types of ground features. There are many methods for remote sensing image classification, which can be divided into supervised classification, unsupervised classification and semi-supervised classification from the perspective of using training samples. Most conventional supervised classification methods start from the spectral characteristics of remote sensing images, and consider the spectral differences between different object types to distinguish pixels. For high-resolution remote sensing images, there is not much spectral information, and usually only includes four bands of RGB and near-infrared, which cannot contain rich spectral information. After classification using spectral classification methods, the classification accuracy is low.

Therefore, how to solve the problem of low classification accuracy of high-resolution remote sensing images is a technical problem to be solved urgently by those skilled in the art.

Contents of the invention

In view of the above problems, the present invention is proposed to provide a semi-supervised classification method, device, equipment and medium for high-resolution remote sensing images that overcome the above problems or at least partially solve the above problems.

A semi-supervised classification method for high-resolution remote sensing images, comprising:

Preprocessing remote sensing images;

According to the remote sensing image after processing, make a classification sample set of ground features;

Construct the remote sensing image semantic segmentation model based on the Unet++ network, and train the semantic segmentation model by the feature classification sample set;

Construct a threshold segmentation model based on the near-infrared band;

performing model fusion on the semantic segmentation model and the threshold segmentation model to obtain a classification model;

The classification model is used to classify the remote sensing images to be classified.

Preferably, in the above semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, the preprocessing of the remote sensing images includes:

Perform panchromatic and multispectral image fusion, radiometric correction, atmospheric correction, and geometric correction processing on remote sensing images.

Preferably, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided in the embodiment of the present invention, the preparation of a ground object classification sample set according to the processed remote sensing images includes:

Establishing a vector of the same size as the processed remote sensing image, and dividing the full elements of the vector into four types: water body, vegetation, bare soil, and impermeable surface;

Label the divided types and convert them into raster data according to the different types of fields;

The processed remote sensing image is sliced, and the marked grid corresponding to the position is cut out at the same time, so as to obtain a group of pictures and labels of the same size as a sample set of ground object classification.

Preferably, in the above-mentioned semi-supervised classification method of high-resolution remote sensing images provided by the embodiment of the present invention, the Unet++ network adopts the method of multi-level upsampling and layer-skip connection to extract multi-layer features.

Preferably, in the above-mentioned semi-supervised classification method of high-resolution remote sensing images provided by the embodiment of the present invention, the Unet++ network includes a downsampling layer, an upsampling layer, and an intermediate layer for feature extraction of the downsampling layer ;in,

The downsampling layer is added with the feature extraction part of the EfficientB4 model.

Preferably, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, the construction of a threshold segmentation model based on the near-infrared band includes:

Using a threshold segmentation method to obtain a threshold histogram of the near-infrared band of the processed remote sensing image;

The ground feature classification sample set is matched with the near-infrared band threshold histogram, and the threshold area of water body and vegetation in the near-infrared band threshold histogram is selected to construct a threshold segmentation model.

Preferably, in the above-mentioned method for semi-supervised classification of high-resolution remote sensing images provided by an embodiment of the present invention, the model fusion of the semantic segmentation model and the threshold segmentation model to obtain a classification model includes:

When the output result of the semantic segmentation model and the output result of the threshold segmentation model are both water bodies or vegetation, or, when the output result of the semantic segmentation model is bare soil or impermeable surface and the threshold segmentation model When the output result is other, it is judged that the classification result is correct;

The classification result determined to be correct is used as a new object classification sample set, and the semantic segmentation model is continuously trained by using the transfer learning method to obtain a classification model.

The embodiment of the present invention also provides a semi-supervised classification device for high-resolution remote sensing images, including:

The image processing module is used for preprocessing the remote sensing images;

The sample set making module is used to make a sample set of ground object classification according to the processed remote sensing image;

The first model building module is used to build a remote sensing image semantic segmentation model based on the Unet++ network, and trains the semantic segmentation model through the feature classification sample set;

The second model building block is used to build a threshold segmentation model based on the near-infrared band;

A model fusion module, configured to perform model fusion on the semantic segmentation model and the threshold segmentation model to obtain a classification model;

An image classification module, configured to use the classification model to classify remote sensing images to be classified.

The embodiment of the present invention also provides a semi-supervised classification device for high-resolution remote sensing images, including a processor and a memory, wherein, when the processor executes the computer program stored in the memory, the above-mentioned A semi-supervised classification method for high-resolution remote sensing images.

An embodiment of the present invention also provides a computer-readable storage medium for storing a computer program, wherein, when the computer program is executed by a processor, the above-mentioned method for semi-supervised classification of high-resolution remote sensing images as provided in the embodiment of the present invention is implemented .

It can be seen from the above technical solutions that a semi-supervised classification method for high-resolution remote sensing images provided by the present invention includes: preprocessing the remote sensing images; making a classification sample set based on the processed remote sensing images; The remote sensing image semantic segmentation model of the Unet++ network, and the semantic segmentation model is trained through the object classification sample set; the threshold segmentation model based on the near-infrared band is constructed; the semantic segmentation model and the threshold segmentation model are model-fused to obtain the classification model; use The classification model classifies the remote sensing images to be classified.

The present invention builds a remote sensing image semantic segmentation model based on the Unet++ network and a threshold segmentation model based on the near-infrared band, and then uses a multi-model fusion method to fuse the texture information of the remote sensing image with the spectral information of the near-infrared band, and then perform high-resolution High-rate remote sensing image classification can improve classification accuracy.

In addition, the present invention also provides corresponding devices, equipment, and computer-readable storage media for the semi-supervised classification method of high-resolution remote sensing images, which further makes the above method more practical. The device, equipment, and computer-readable storage media have corresponding The advantages.

The above description is only an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the description, and in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable , the specific embodiments of the present invention are enumerated below.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiment. The drawings are only for the purpose of illustrating a preferred embodiment and are not to be considered as limiting the invention. Also throughout the drawings, the same reference numerals are used to designate the same components. In the attached picture:

Fig. 1 shows the flow chart of the semi-supervised classification method of high-resolution remote sensing image provided by the embodiment of the present invention;

2 shows a schematic diagram of a semi-supervised classification method for high-resolution remote sensing images provided by an embodiment of the present invention;

Fig. 3 shows the structural representation of the Unet++ network that the embodiment of the present invention provides;

Fig. 4 shows the schematic structural diagram of the convolution block in the Unet++ network provided by the embodiment of the present invention;

Fig. 5 shows a schematic structural diagram of the residual block in the Unet++ network provided by the embodiment of the present invention;

Fig. 6 shows the near-infrared band threshold histogram provided by the embodiment of the present invention;

FIG. 7 shows a display diagram of classification results provided by an embodiment of the present invention;

Fig. 8 shows a schematic structural diagram of an apparatus for semi-supervised classification of high-resolution remote sensing images provided by an embodiment of the present invention.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The present invention provides a semi-supervised classification method for high-resolution remote sensing images, as shown in Figure 1 and Figure 2, comprising the following steps:

S101. Preprocessing the remote sensing images;

S102. According to the processed remote sensing image, make a sample set for ground object classification;

S103. Construct a remote sensing image semantic segmentation model based on the Unet++ network, and train the semantic segmentation model through the sample set of ground object classification;

S104. Construct a threshold segmentation model based on the near-infrared band;

S105. Perform model fusion of the semantic segmentation model and the threshold segmentation model to obtain a classification model;

S106. Classify the remote sensing image to be classified by using the classification model.

In the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, by constructing a remote sensing image semantic segmentation model based on the Unet++ network and a threshold segmentation model based on the near-infrared band, and then using a multi-model fusion method, the remote sensing The texture information of the image and the spectral information of the near-infrared band are fused, and then the classification of high-resolution remote sensing images can improve the classification accuracy.

Further, in specific implementation, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, step S101 performs preprocessing on remote sensing images, which may specifically include: performing panchromatic and multispectral image processing on remote sensing images Fusion, radiation correction, atmospheric correction, geometric correction and other processing to obtain a high-resolution remote sensing image with four bands (RGB and near-infrared).

During specific implementation, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, since the size of the remote sensing images is too large, it needs to be cut into slices and given attributes. Step S102 is based on the processed remote sensing images. Object classification sample set, which may specifically include:

First, establish a vector with the same size as the processed remote sensing image, and divide the full elements of the vector into four types: water body, vegetation, bare soil, and impermeable surface;

Then, mark the divided types and convert them into raster data according to the different type fields; specifically, you can use the shpfile file to mark, and then convert the shpfile into raster data according to the different type fields;

Finally, the processed remote sensing image is sliced; specifically, the remote sensing image is cut into small slices for training. The slice size can be set to 512×512, and the slices do not overlap; at the same time, the label of the corresponding position is cut out A set of images and labels of the same size is obtained as a sample set for object classification.

During specific implementation, in the above-mentioned semi-supervised classification method of high-resolution remote sensing images provided by the embodiment of the present invention, as shown in Figure 3, the Unet++ network adopts the method of multi-level upsampling and layer-skip connection to extract multi-layer features. It should be noted that the Unet network is a commonly used segmentation network model in semantic segmentation. First, downsampling is performed through convolution, layer after layer features are extracted, and then upsampling is performed, and the difference between downsampling and upsampling is The feature connection of the model, because the shape of this model structure diagram is similar to U-shaped, so it is named Unet, where the process of downsampling is an encoding process, and upsampling is a process of decoding; while the Unet++ network is based on Unet. The method of multi-level upsampling and layer-skip connection is adopted to extract features of more layers.

In specific implementation, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, the Unet++ network can specifically include a downsampling layer, an upsampling layer, and an intermediate layer for feature extraction of the downsampling layer ; Among them, the feature extraction part of the EfficientB4 model is added to the downsampling layer. That is to say, on the basis of Unet, the present invention adds the feature extraction part of the EfficientB4 model into the encoding process of the Unet++ network, improves the network structure, and extracts more features.

Specifically, step S103 builds a remote sensing image semantic segmentation model based on the Unet++ network, which may specifically include the following steps:

First, define a convolution block, as shown in Figure 4, a convolution block includes a convolution layer, BatchNormalization (BN) layer, and a LeakyRelU activation function;

Then, define a residual block, as shown in Figure 5, after the input of the residual block network passes through the LeakReLU and BN layers, then through the two convolution blocks defined above, and then add the original input after the BN layer processing result;

After that, the downsampling layer, upper adoption layer and middle layer of the Unet++ network are established. The downsampling layer is the part of feature extraction. It is the same as the Unet network. Here, four downsampling layers are used as the downsampling layer of the Unet++ network. The difference is that in order to propose features more deeply, the downsampling layer of the network is obtained from EfficientB4 , that is, the 342nd, 154th, 92nd, and 30th layers of EfficientB4 are used as the four layers conv4, conv3, conv2, and conv1 of the downsampling in Unet++; the middle layer is a further feature extraction for the downsampling layer, and the features of different levels are not given. , and the number of extractions is also different. For conv4, it is coded as deconv4, and then extracted three times to obtain the three-level feature layer deconv4_up1, deconv4_up2, deconv4_up3, then extract the features of conv4 as deconv3, and extract the features deconv3_up1, deconv3_up2, add deconv3, conv3 and deconv4_up1 to obtain uconv3, the Uconv3 is encoded as deconv2, and the feature deconv2_up1 is extracted, and then deconv2, conv2, deconv4_up2, deconv3_up1 are added to obtain uconv2, uconv2 is encoded as deconv1, and then conv1, deconv1, deconv2_up1, deconv3_up2, deconv4_up3 are added to obtain uconv1, and encoded It is uconv0, and finally make a convolution of uconv0, reduce the feature to 1 dimension, and use it as the network output;

Finally, define the loss function, using the dice loss between the network prediction result and the real label as the loss function. Use the cosine annealing method to set the learning rate (lr), the initial lr is set to 0.001, the epoch is set to 30, the batchsize is set to 32, and then the training model is started.

In specific implementation, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, step S104 constructs a threshold segmentation model based on the near-infrared band, which may specifically include: using the threshold segmentation method to obtain processed remote sensing images The threshold histogram of the near-infrared band; match the object classification sample set with the threshold histogram of the near-infrared band, and select the threshold area of water and vegetation in the threshold histogram of the near-infrared band to build a threshold segmentation model.

Specifically, considering that the near-infrared band has a strong absorption of water bodies, and vegetation contains more water bodies, so the reflectivity of water bodies and vegetation is low, band 4 is used to judge water bodies and vegetation. In step S104, the method of threshold segmentation is used to extract only band 4 in the high-resolution remote sensing image, and display the histogram. For 8-bit images, the abscissa of the histogram ranges from 0 to 255, as shown in Figure 6. The training samples and The histogram is matched, and a section of the water body in the histogram is selected as the reflectance part of the water body in the image, usually on the left, to record the threshold area of the water body. Similarly, the threshold area of the vegetation is selected. When the data to be classified is within the threshold of water body or vegetation, it is marked as water body or vegetation. In this way, making full use of the characteristics of the near-infrared band can increase the classification accuracy of specific targets to a certain extent.

In specific implementation, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, step S105 performs model fusion of the semantic segmentation model and the threshold segmentation model to obtain the classification model, which may specifically include: when the semantic segmentation model When the output result of the threshold segmentation model and the output result of the threshold segmentation model are both water bodies or vegetation, or when the output result of the semantic segmentation model is bare soil or impermeable surface and the output result of the threshold segmentation model is other, it is determined that the classification result is correct; The classification result judged to be correct is used as a new object classification sample set, and the transfer learning method is used to continue training the semantic segmentation model to obtain the classification model. In this way, semantic segmentation is used to classify remote sensing images, and at the same time, for some types of ground objects such as water bodies, vegetation, etc., the near-infrared band features are used to extract, and the two results are fused at the decision level, and then the semi-supervised classification method is used to increase samples, and Classification again can obtain higher classification accuracy.

Specifically, step S105 uses a decision-level fusion method. The output result of the semantic segmentation model is a two-dimensional matrix, and each position represents the category of the pixel, including four categories in total. The result of threshold segmentation is also a two-dimensional matrix. Each location represents the category of the pixel. Unlike semantic segmentation, it only contains three categories, water body, vegetation, and others. For each pixel, when the result of semantic segmentation is bare soil or impermeable surface, the result of threshold segmentation is other, or both are water bodies or vegetation, the classification is considered correct. When the classification results are different, this part is not classified, and the classified one is used as a new training sample, and the transfer learning method is used to continue training the semantic segmentation model to obtain the final classification result.

During specific implementation, in the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention, step S106 uses the classification model to classify the remote sensing images to be classified, which may specifically include: first, slice the remote sensing images to be classified, slice The size is the same as step 1, and then each slice is input into the classification model for classification, after expansion, corrosion and other post-processing operations, and then the classification results of the last slices are stitched together, coordinate information is added, and a tiff file is generated, an example is shown in Figure 7 shown.

It should be noted that the above-mentioned semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention mainly considers the texture features of high-resolution remote sensing images and the physical characteristics of the near-infrared band, and uses the improved Unet++ semantic segmentation network and threshold segmentation In a combined way, the semi-supervised classification method is used to expand the training samples, and the semantic segmentation model is trained again to obtain higher classification accuracy, and finally the classification results are generated after classification.

Based on the same inventive concept, an embodiment of the present invention also provides a semi-supervised classification device for high-resolution remote sensing images. Since the problem-solving principle of the device is similar to the aforementioned semi-supervised classification method for high-resolution remote sensing images, the device's For the implementation, please refer to the implementation of the semi-supervised classification method for high-resolution remote sensing images, and the repetition will not be repeated.

In specific implementation, the high-resolution remote sensing image semi-supervised classification device provided by the embodiment of the present invention, as shown in Figure 8, specifically includes:

An image processing module 11, configured to preprocess remote sensing images;

A sample set making module 12, configured to make a sample set for classification of features according to the processed remote sensing images;

The first model building module 13 is used to build a remote sensing image semantic segmentation model based on the Unet++ network, and trains the semantic segmentation model through the feature classification sample set;

The second model construction module 14 is used to construct a threshold segmentation model based on the near-infrared band;

The model fusion module 15 is used to carry out model fusion with the semantic segmentation model and the threshold segmentation model to obtain the classification model;

The image classification module 16 is configured to use a classification model to classify the remote sensing image to be classified.

In the above-mentioned semi-supervised classification device for high-resolution remote sensing images provided by the embodiment of the present invention, the texture information of the remote sensing images and the spectral information in the near-infrared band can be fused through the interaction of the above six modules, and then the high-resolution Remote sensing image classification improves classification accuracy.

For the more specific working process of each of the above modules, reference may be made to the corresponding content disclosed in the foregoing embodiments, which will not be repeated here.

Correspondingly, the embodiment of the present invention also discloses a semi-supervised classification device for high-resolution remote sensing images, including a processor and a memory; wherein, when the processor executes the computer program stored in the memory, the high-resolution remote sensing image disclosed in the foregoing embodiments is realized. Image semi-supervised classification method.

For a more specific process of the above method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details are not repeated here.

Further, the present invention also discloses a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the aforementioned semi-supervised classification method for high-resolution remote sensing images is realized.

For a more specific process of the above method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details are not repeated here.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same or similar parts of each embodiment can be referred to each other. For the devices, equipment, and storage media disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and for relevant details, please refer to the description of the methods.

Professionals can further realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software or a combination of the two. In order to clearly illustrate the possible For interchangeability, in the above description, the composition and steps of each example have been generally described according to their functions. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present application.

The steps of the methods or algorithms described in connection with the embodiments disclosed herein may be directly implemented by hardware, software modules executed by a processor, or a combination of both. Software modules can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other Any other known storage medium.

In summary, a semi-supervised classification method for high-resolution remote sensing images provided by the embodiment of the present invention includes: preprocessing the remote sensing images; making a sample set of ground object classification according to the processed remote sensing images; constructing a remote sensing image based on the Unet++ network Image semantic segmentation model, and train the semantic segmentation model through the object classification sample set; build a threshold segmentation model based on the near-infrared band; fuse the semantic segmentation model and the threshold segmentation model to obtain a classification model; use the classification model to treat classification classification of remote sensing images. In this way, by constructing a remote sensing image semantic segmentation model based on the Unet++ network and a threshold segmentation model based on the near-infrared band, and then using a multi-model fusion method, the texture information of the remote sensing image and the spectral information of the near-infrared band are fused, and then the high-resolution Remote sensing image classification can improve classification accuracy. In addition, the present invention also provides corresponding devices, equipment, and computer-readable storage media for the semi-supervised classification method of high-resolution remote sensing images, which further makes the above method more practical. The device, equipment, and computer-readable storage media have corresponding The advantages.

Finally, it should also be noted that in this text, relational terms such as first and second etc. are only used to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations, any such actual relationship or order exists. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

The method, device, equipment and medium for the semi-supervised classification of high-resolution remote sensing images provided by the present invention have been introduced in detail above. In this paper, specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only It is used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation and scope of application. In summary, this The content of the description should not be construed as limiting the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

Claims (10)

A semi-supervised classification method for high-resolution remote sensing images, characterized in that it includes: Preprocessing remote sensing images; According to the remote sensing image after processing, make a classification sample set of ground features; Construct the remote sensing image semantic segmentation model based on the Unet++ network, and train the semantic segmentation model by the feature classification sample set; Construct a threshold segmentation model based on the near-infrared band; performing model fusion on the semantic segmentation model and the threshold segmentation model to obtain a classification model; The classification model is used to classify the remote sensing images to be classified. The semi-supervised classification method for high-resolution remote sensing images according to claim 1, wherein said preprocessing the remote sensing images includes: Perform panchromatic and multispectral image fusion, radiometric correction, atmospheric correction, and geometric correction processing on remote sensing images. The method for semi-supervised classification of high-resolution remote sensing images according to claim 2, wherein, according to the processed remote sensing images, making a classification sample set of ground features includes: Establishing a vector of the same size as the processed remote sensing image, and dividing the full elements of the vector into four types: water body, vegetation, bare soil, and impermeable surface; Label the divided types and convert them into raster data according to the different types of fields; The processed remote sensing image is sliced, and the marked grid corresponding to the position is cut out at the same time, so as to obtain a group of pictures and labels of the same size as a sample set of ground object classification. The semi-supervised classification method for high-resolution remote sensing images according to claim 3, wherein the Unet++ network adopts multi-level up-sampling and layer-skip connection methods to extract multi-layer features. The semi-supervised classification method for high-resolution remote sensing images according to claim 4, wherein the Unet++ network includes a downsampling layer, an upsampling layer, and an intermediate layer for feature extraction to the downsampling layer; in, The downsampling layer is added with the feature extraction part of the EfficientB4 model. The semi-supervised classification method for high-resolution remote sensing images according to claim 5, wherein said building a threshold segmentation model based on near-infrared bands includes: Using a threshold segmentation method to obtain a threshold histogram of the near-infrared band of the processed remote sensing image; The ground object classification sample set is matched with the near-infrared band threshold histogram, and the threshold area of water body and vegetation in the near-infrared band threshold histogram is selected to construct a threshold segmentation model. The semi-supervised classification method for high-resolution remote sensing images according to claim 6, wherein the model fusion of the semantic segmentation model and the threshold segmentation model to obtain a classification model includes: When the output result of the semantic segmentation model and the output result of the threshold segmentation model are both water bodies or vegetation, or, when the output result of the semantic segmentation model is bare soil or impermeable surface and the threshold segmentation model When the output result is other, it is judged that the classification result is correct; The classification result determined to be correct is used as a new object classification sample set, and the semantic segmentation model is continuously trained by using the transfer learning method to obtain a classification model. A semi-supervised classification device for high-resolution remote sensing images, characterized in that it includes: The image processing module is used for preprocessing the remote sensing images; The sample set making module is used to make a sample set of ground object classification according to the processed remote sensing image; The first model building module is used to build a remote sensing image semantic segmentation model based on the Unet++ network, and trains the semantic segmentation model through the feature classification sample set; The second model building block is used to build a threshold segmentation model based on the near-infrared band; A model fusion module, configured to perform model fusion on the semantic segmentation model and the threshold segmentation model to obtain a classification model; An image classification module, configured to use the classification model to classify remote sensing images to be classified. A semi-supervised classification device for high-resolution remote sensing images, characterized in that it includes a processor and a memory, wherein when the processor executes the computer program stored in the memory, it realizes any one of claims 1 to 7 A semi-supervised classification method for high-resolution remote sensing images. A computer-readable storage medium, characterized in that it is used to store a computer program, wherein, when the computer program is executed by a processor, the semi-supervised classification of high-resolution remote sensing images according to any one of claims 1 to 7 is realized method.

Images