Image processing method, system, device and computer medium

Abstract

The present invention discloses an image processing method, system, device and computer medium. The method comprises: obtaining an image to be processed; calling an image segmentation model; wherein the encoder in the image segmentation model comprises an N-layer downsampling network, and the N-layer downsampling network in the image segmentation model performs N-layer feature extraction on the image to be processed to obtain a target downsampling feature map; the decoder in the image segmentation model performs decoding processing on the target downsampling feature map to obtain a segmentation result of the image to be processed. The above method fuses the feature map extracted during encoding with the input feature map by utilizing the skip connection characteristics of the dense connection module, thereby effectively retaining the information transmission between layers to improve the segmentation accuracy of the processing result.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image processing method, system, device and computer medium. Background Art

[0002] In the field of image processing, image segmentation is a processing technology that divides an image into several regions with similar properties. It divides the image into several non-overlapping regions based on features such as grayscale, color, spatial texture, and geometric shape, so that these features show consistency or similarity in the same region and show obvious differences between different regions, thereby separating the target from the background.

[0003] In the prior art, threshold segmentation is a commonly used image segmentation algorithm. The basic idea of ​​this method is to calculate one or more grayscale thresholds based on the grayscale features of the image, compare the grayscale value of each pixel in the image with the threshold, and finally classify the pixels into appropriate categories based on the comparison results.

[0004] However, this method usually only considers the characteristics of the pixel grayscale value itself, but not the spatial characteristics. Therefore, it is sensitive to noise and the segmentation result accuracy is poor. Summary of the invention

[0005] In view of the above technical problems, the present invention provides an image processing method, system, device and computer medium, which can effectively improve the segmentation accuracy of image processing results.

[0006] In a first aspect, the present invention provides an image processing method, comprising:

[0007] Get the image to be processed;

[0008] Calling an image segmentation model; wherein the encoder in the image segmentation model includes N layers of downsampling networks, each of which includes a convolution module, a first pooling module, a second pooling module, and a dense connection module connected to the first pooling module and the second pooling module;

[0009] The image to be processed is subjected to N-layer feature extraction through the N-layer downsampling network in the image segmentation model to obtain a target downsampling feature map; wherein N is a non-zero natural number; in each layer feature extraction process:

[0010] The convolution module and the first pooling module of the current layer are used in sequence to process the downsampled feature map output by the previous layer to obtain a first feature map; the second pooling module of the current layer is used to perform pooling processing on the image to be processed to obtain a second feature map; the dense connection module of the current layer is used to densely connect the first feature map and the second feature map to obtain a downsampled feature map of the current layer;

[0011] The target downsampled feature map is decoded by a decoder in the image segmentation model to obtain a segmentation result of the image to be processed.

[0012] Optionally, the performing N-layer feature extraction on the image to be processed through an N-layer downsampling network in the image segmentation model further includes:

[0013] If the current layer is the first layer: sequentially use the convolution module and the first pooling module of the current layer to process the image to be processed to obtain a first feature map; use the second pooling module of the current layer to perform pooling processing on the image to be processed to obtain a second feature map;

[0014] The first feature map and the second feature map are densely connected using the dense connection module of the current layer to obtain a downsampled feature map of the current layer.

[0015] Optionally, the decoder in the image segmentation model includes N layers of upsampling networks, each of which includes a convolution module and a cropping and splicing module;

[0016] The upsampling network of the Mth layer and the downsampling network of the (N-M+1)th layer are connected through the cropping and splicing module; M is less than N and M is a non-zero natural number.

[0017] Optionally, the decoding process of the target down-sampled feature map by a decoder in the image segmentation model to obtain a segmentation result of the image to be processed is specifically:

[0018] The target down-sampled feature map is subjected to N-layer up-sampling processing through the N-layer up-sampling network to obtain a target up-sampling result; in each layer of up-sampling processing:

[0019] Using the convolution module of the current layer to process the upsampled feature map output by the previous layer to obtain a third feature map;

[0020] The cropping and splicing module is used to crop and splice the down-sampled feature map output by the down-sampling network connected to the current layer and the third feature map to obtain the up-sampled feature map of the current layer.

[0021] Optionally, the cropping and splicing module is used to crop and splice the down-sampled feature map output by the down-sampling network connected to the current layer and the third feature map, specifically:

[0022] The down-sampled feature map output by the down-sampling network connected to the current layer is cropped by the cropping and splicing module, and the cropped down-sampled feature map is spliced ​​with the third feature map to obtain the up-sampled feature map of the current layer.

[0023] Optionally, performing N-layer upsampling processing on the target downsampled feature map through the N-layer upsampling network further includes:

[0024] If the current layer is the first layer: using the convolution module of the current layer to process the target down-sampled feature map output by the encoder in the image segmentation model to obtain a third feature map;

[0025] The cropping and splicing module is used to crop and splice the down-sampled feature map output by the down-sampling network connected to the current layer and the third feature map to obtain the up-sampled feature map of the current layer.

[0026] Optionally, performing N-layer upsampling processing on the target downsampled feature map through the N-layer upsampling network further includes:

[0027] If the current layer is the Nth layer: using the convolution module of the current layer to process the upsampled feature map output by the previous layer to obtain a third feature map;

[0028] The third feature map is padded with zeros based on the size of the image to be processed to obtain an upsampled feature map of the current layer.

[0029] In a second aspect, the present invention provides an image processing system, comprising:

[0030] An acquisition unit, used for acquiring an image to be processed;

[0031] The encoding unit is used to call the image segmentation model; wherein the encoder in the image segmentation model includes N layers of downsampling networks, each of which includes a convolution module, a first pooling module, a second pooling module, and a dense connection module connected to the first pooling module and the second pooling module; the N layers of downsampling networks in the image segmentation model are used to extract N layers of features from the image to be processed to obtain a target downsampling feature map; wherein N is a non-zero natural number; in each layer of feature extraction process:

[0032] The convolution module and the first pooling module of the current layer are used in sequence to process the downsampled feature map output by the previous layer to obtain a first feature map; the second pooling module of the current layer is used to perform pooling processing on the image to be processed to obtain a second feature map; the dense connection module of the current layer is used to densely connect the first feature map and the second feature map to obtain a downsampled feature map of the current layer;

[0033] A decoding unit is used to decode the target down-sampled feature map through a decoder in the image segmentation model to obtain a segmentation result of the image to be processed.

[0034] In a third aspect, the present invention provides a data processing device, comprising a processor, wherein the processor is coupled to a memory, the memory stores a program, and the program is executed by the processor, so that the data processing device executes the image processing method described in the first aspect above.

[0035] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the image processing method as described in the first aspect above is implemented.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The image processing method provided by the present invention, in the process of encoding the image to be processed, fuses the feature map extracted during encoding with the input feature map by utilizing the skip connection characteristics of the dense connection module, effectively retains the information transmission between layers, thereby improving the segmentation accuracy of the processing result. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the implementation mode will be briefly introduced below. Obviously, the drawings described below are only some implementation modes of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 is a flowchart of an image processing method provided by an embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of the structure of an image processing system provided by an embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of the operation of cutting and splicing layers provided by an embodiment of the present invention;

[0042] Figure 4 It is a structural block diagram of an image segmentation model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] See also Figure 1In a first aspect, an embodiment of the present invention provides an image processing method, which specifically includes the following S1 to S3.

[0045] S1: Obtain the image to be processed.

[0046] S2: calling an image segmentation model; wherein the encoder in the image segmentation model includes an N-layer downsampling network; performing N-layer feature extraction on the image to be processed through the N-layer downsampling network in the image segmentation model to obtain a target downsampling feature map, wherein N is a non-zero natural number. Specifically, N can be 4.

[0047] It should be noted that each downsampling network includes a convolution module, a first pooling module, a second pooling module, and a dense connection module connected to the first pooling module and the second pooling module.

[0048] In the feature extraction process of each layer: the convolution module and the first pooling module of the current layer are used in turn to process the downsampled feature map output by the previous layer to obtain a first feature map; the second pooling module of the current layer is used to perform pooling processing on the image to be processed to obtain a second feature map; the dense connection module of the current layer is used to densely connect the first feature map and the second feature map to obtain the downsampled feature map of the current layer.

[0049] Specifically, if the current layer is the first layer, the convolution module and the first pooling module of the current layer are used in sequence to process the image to be processed to obtain a first feature map; the second pooling module of the current layer is used to perform pooling processing on the image to be processed to obtain a second feature map; the dense connection module of the current layer is used to densely connect the first feature map and the second feature map to obtain a downsampled feature map of the current layer.

[0050] S3: Decoding the target downsampled feature map through the decoder in the image segmentation model to obtain a segmentation result of the image to be processed.

[0051] It should be noted that the decoder in the image segmentation model includes N layers of upsampling networks, each of which includes a convolution module and a cropping and splicing module; the Mth layer of the upsampling network and the (N-M+1)th layer of the downsampling network are connected through the cropping and splicing module; M is less than N and M is a non-zero natural number.

[0052] Exemplarily, when the number of layers N of the downsampling network and the upsampling network is 4, the 3rd layer of the upsampling network in the decoder is connected to the 2nd layer of the downsampling network in the encoder.

[0053] Correspondingly, when the target down-sampled feature map is subjected to N-layer up-sampling processing through the N-layer up-sampling network, in each layer of up-sampling processing: the up-sampled feature map output by the previous layer is processed by the convolution module of the current layer to obtain a third feature map; the down-sampled feature map output by the down-sampling network connected to the current layer is cropped and spliced ​​with the third feature map by the cropping and splicing module to obtain the up-sampled feature map of the current layer.

[0054] It should be noted that if the current layer is the first layer, the convolution module of the current layer is used to process the target down-sampled feature map output by the encoder in the image segmentation model to obtain a third feature map; the cropping and splicing module is used to crop and splice the down-sampled feature map output by the down-sampling network connected to the current layer and the third feature map to obtain the up-sampled feature map of the current layer.

[0055] It should be noted that if the current layer is the Nth layer, the convolution module of the current layer is used to process the upsampled feature map output by the previous layer to obtain a third feature map; and then the third feature map is padded with zeros based on the size of the image to be processed to obtain the upsampled feature map of the current layer.

[0056] Specifically, in this embodiment, after the up-sampled feature map is outputted through the Nth layer, the up-sampled feature map is binarized based on a preset threshold value to generate a segmentation result corresponding to the image to be processed.

[0057] The image processing method provided by the above embodiment of the present invention, during the process of encoding the image to be processed, fuses the feature map extracted during encoding with the input feature map by utilizing the skip connection characteristics of the densely connected module, thereby effectively retaining the information transmission between layers, thereby improving the segmentation accuracy of the processing result.

[0058] The implementation process of the image processing method will be described below through a specific embodiment.

[0059] In this embodiment, the image segmentation model called by the execution process of image processing is a pre-trained udensedown network model.

[0060] See also Figure 2The udensedown network model includes an input layer (Input), 20 convolutional layers (C1, C2, ..., C20), 4 densely connected layers (Dense1, Dense2, Dense3 and Dense4), 4 pooling layers (P1, P2, P3 and P4), 4 upsampling layers (U1, U2, U3 and U4) and 4 cropping and splicing layers (Crop1+Conc1, Crop2+Conc2, Crop3+Conc3 and Crop4+Conc4); among them, the input layer is used as the input end of the model to convert the preprocessed image to be processed into a vector representation; and the first 5 convolutional layers of the model are used to extract image features more fully and finely, and have a larger receptive field.

[0061] The udensedown network model is improved and constructed based on the classic image segmentation model U-Net. The original U-Net network has a U-shaped structure, and the entire network is an encoder-decoder structure. The entire network reuses two cascaded convolutional layers in large quantities. Each convolutional layer uses a 3*3 convolution kernel, followed by a rectified linear unit (ReLU).

[0062] The encoding process of the U-Net network consists of four steps, each of which performs convolution and pooling operations to extract image features and halve the size of the image. The decoding process of the network is similar to the encoding process, specifically replacing the maximum pooling operation with an upsampling operation, which doubles the size of the input while restoring the resolution of the output image to the original input image size.

[0063] Since images have different sizes, in order to more fully extract image features, this embodiment constructs a udensedown network model based on U-Net.

[0064] In the encoding stage of the udensedown network model, the input image to be processed is first subjected to continuous convolution operations through the convolution layers C1, C2 and C3. It should be noted that the convolution kernel size of the convolution layers C1, C2 and C3 is 5*5 to extract richer image features.

[0065] Compared with U-Net, the encoder of the udensedown network model also adds a densely connected layer to ensure that the input of each layer is a collection of the input layer and the output of the previous layer, so that the encoder retains the information transmission between layers and alleviates the gradient vanishing problem caused by the network being too deep.

[0066] For example, when the input image to be processed is a 200*200 image dataset, the input layer of the udensedown network model receives a 200*200*1 tensor.

[0067] In the encoding stage of the udensedown network model, the input image is firstly subjected to continuous convolution operations based on three layers of convolutional layers (C1, C2 and C3) with a convolution kernel of 5*5, and 64 feature maps of 200*200 are obtained, that is, the input matrix dimension of the first pooling layer (P1) is 200*200*64. Furthermore, after the maximum pooling (P1), a feature map of 100*100*64 is obtained (denoted as F1).

[0068] At the same time, based on the dense connection layer, the 200*200*1 input by the input layer is directly pooled (P5) to obtain a 100*100*1 feature map (denoted as F2), and then the feature map F2 is spliced ​​with the feature map F1 to obtain a 100*100*65 feature map as the input of the fourth convolutional layer (C4). It can be understood that the dense connection layer set in this embodiment is essentially a set of continuous operations of pooling-splicing.

[0069] Specifically, after the input layer undergoes the maximum pooling operation in the pooling layer (P5), its size is reduced to 100*100, and then it is concatenated with the output features of P1 to form a feature map of 100*100*65, which serves as the input of the C5 layer.

[0070] Compared with the encoding process of the U-Net network, the input features of the convolutional layer of the udensedown network model include not only the feature map information from the output of the previous convolutional layer, but also the information from the input layer, which can reflect the important reference role of the input image for feature extraction.

[0071] After the above four rounds of convolution-pooling-concatenation operations, the target feature map output by the encoder has a size of 12*12.

[0072] In order to restore the resolution of the image, during the decoding process, this embodiment uses upsampling layers (U1, U2, U3 and U4) with a size of 2*2 to gradually restore the resolution of the target feature map output by the encoder. For example, after upsampling at the U1 layer, the size of the feature map is restored from 12*12 to 24*24.

[0073] Specifically, in the decoding stage, this embodiment also sets a cropping and splicing layer to combine the output features of each layer in the decoding process with the high-resolution features of the corresponding layer in the encoding process. For example, the Crop1+Conc1 layer crops the 25*25*256 feature map output by the C9 layer to 24*24*256, and then splices it with the 24*24*512 feature map obtained by the U1 layer to obtain a 24*24*768 feature map.

[0074] It should be noted that the specific operation process of cutting and splicing layers is as follows Figure 3 For example, the feature map output by the convolution layer C3 is 200*200*64 (denoted as F3), and the feature map output by the U4 layer is 192*192*96 (denoted as F4). The size of F3 is first cropped from 200*200 to 192*192, and then the feature map of 192*192*64 is concatenated with the feature map of 192*192*96 to obtain a feature map of 192*192*160.

[0075] Specifically, in order to obtain more accurate output features, this embodiment also performs two consecutive convolution operations (C12 and C13) on the feature results obtained by the cropped and spliced ​​layers.

[0076] This embodiment can combine the context information output by the shallow network during the encoding process with the relatively abstract semantic information output by the deep network during the decoding process by using the cropped splicing layer, so as to provide richer information for the convolution operation.

[0077] In this embodiment, after repeating the above upsampling-cropping-splicing-convolution operation four times, a feature map of 196*196*64 can be obtained, that is, the output of the convolution layer C19 is a feature map of 196*196*64.

[0078] After the output feature map is obtained by the convolution layer C19, the output feature map of C19 is padded with zeros to expand the size of the output feature map to 200*200.

[0079] In this embodiment, the last convolution layer C20 is the output layer of the udensedown network model. This embodiment uses a 1*1 convolution kernel to perform a convolution operation, and then obtains a 200*200*1 feature map through a sigmoid activation function, which is the output of the udensedown network model.

[0080] It can be understood that after each pixel in the final output feature map passes through the sigmoid activation function, the result is mapped to 0 to 1, thereby obtaining the final segmentation prediction probability map.

[0081] The present invention will describe the training process of the udensedown network model through another embodiment, and the specific process is as follows:

[0082] 1) Build the udensedown network model architecture based on the U-Net algorithm.

[0083] 2) Obtain a training image set and a prediction image set, including original images and corresponding gold standard images.

[0084] 3) Preprocess the training image set and the prediction image set, including: unifying the image size through cropping or padding operations, normalizing the grayscale value of the image, etc.

[0085] 4) Perform data enhancement on the preprocessed training images and their corresponding gold standard images, including flipping, scaling, rotating and shearing.

[0086] 5) The model is trained using the data-enhanced training image set. By reasonably setting hyperparameters, using the back propagation algorithm and the Adam optimization algorithm, a fixed-parameter udensedown network model is obtained.

[0087] 6) Use the predicted image set to predict the trained udensedown network model and obtain the segmentation probability map.

[0088] 7) Set a threshold based on experience, and perform a binarization operation on the segmentation probability map based on the threshold to obtain a segmentation prediction map.

[0089] Optionally, the threshold may be set to 0.5.

[0090] See also Figure 4 In a second aspect, an embodiment of the present invention further provides an image processing system, which specifically includes an acquisition unit 101, an encoding unit 102 and a decoding unit 103.

[0091] The acquisition unit 101 is used to acquire an image to be processed.

[0092] The encoding unit 102 is used to call an image segmentation model; wherein the encoder in the image segmentation model includes an N-layer downsampling network, each of the downsampling networks includes a convolution module, a first pooling module, a second pooling module, and a dense connection module connected to the first pooling module and the second pooling module; the N-layer downsampling network in the image segmentation model is used to perform N-layer feature extraction on the image to be processed to obtain a target downsampling feature map; wherein N is a non-zero natural number.

[0093] In the feature extraction process of each layer: the convolution module and the first pooling module of the current layer are used in turn to process the downsampled feature map output by the previous layer to obtain a first feature map; the second pooling module of the current layer is used to perform pooling processing on the image to be processed to obtain a second feature map; the dense connection module of the current layer is used to densely connect the first feature map and the second feature map to obtain the downsampled feature map of the current layer.

[0094] The decoding unit 103 is used to decode the target down-sampled feature map through a decoder in the image segmentation model to obtain a segmentation result of the image to be processed.

[0095] The information interaction, execution process, etc. between the units in the above-mentioned system are based on the same concept as the image processing method embodiment described in the first aspect of the present invention. More specific contents can be found in the description of the method embodiment of the present invention and will not be repeated here.

[0096] In a third aspect, the present invention provides a data processing device, comprising a processor, wherein the processor is coupled to a memory, the memory stores a program, and the program is executed by the processor, so that the data processing device executes the image processing method described in the first aspect of the present invention.

[0097] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the image processing method as described in the first aspect of the present invention is implemented.

[0098] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-monitorable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0099] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An image processing method, characterized in that: include: Get the image to be processed; Calling an image segmentation model; wherein the encoder in the image segmentation model includes N layers of downsampling networks, each of which includes a convolution module, a first pooling module, a second pooling module, and a dense connection module connected to the first pooling module and the second pooling module; The image to be processed is subjected to N-layer feature extraction through the N-layer downsampling network in the image segmentation model to obtain a target downsampling feature map; wherein N is a non-zero natural number; in each layer feature extraction process: The convolution module and the first pooling module of the current layer are used in sequence to process the downsampled feature map output by the previous layer to obtain a first feature map; the second pooling module of the current layer is used to perform pooling processing on the image to be processed to obtain a second feature map; the dense connection module of the current layer is used to densely connect the first feature map and the second feature map to obtain a downsampled feature map of the current layer; The target downsampled feature map is decoded by a decoder in the image segmentation model to obtain a segmentation result of the image to be processed.

2. The image processing method according to claim 1, characterized in that: The performing N-layer feature extraction on the image to be processed by the N-layer downsampling network in the image segmentation model also includes: If the current layer is the first layer: sequentially use the convolution module and the first pooling module of the current layer to process the image to be processed to obtain a first feature map; use the second pooling module of the current layer to perform pooling processing on the image to be processed to obtain a second feature map; The first feature map and the second feature map are densely connected using the dense connection module of the current layer to obtain a downsampled feature map of the current layer.

3. The image processing method according to claim 1, characterized in that: The decoder in the image segmentation model includes N layers of upsampling networks, each of which includes a convolution module and a cropping and splicing module; The upsampling network of the Mth layer and the downsampling network of the (N-M+1)th layer are connected through the cropping and splicing module; M is less than N and is a non-zero natural number.

4. The image processing method according to claim 3, characterized in that: The decoding process of the target down-sampled feature map by the decoder in the image segmentation model to obtain the segmentation result of the image to be processed is specifically as follows: The target down-sampled feature map is subjected to N-layer up-sampling processing through the N-layer up-sampling network to obtain a target up-sampling result; in each layer of up-sampling processing: Using the convolution module of the current layer to process the upsampled feature map output by the previous layer to obtain a third feature map; The cropping and splicing module is used to crop and splice the down-sampled feature map output by the down-sampling network connected to the current layer and the third feature map to obtain the up-sampled feature map of the current layer.

5. The image processing method according to claim 4, characterized in that: The cropping and splicing module is used to crop and splice the down-sampled feature map output by the down-sampling network connected to the current layer and the third feature map, specifically: The down-sampled feature map output by the down-sampling network connected to the current layer is cropped by the cropping and splicing module, and the cropped down-sampled feature map is spliced ​​with the third feature map to obtain the up-sampled feature map of the current layer.

6. The image processing method according to claim 4, characterized in that: The performing N-layer upsampling processing on the target down-sampled feature map through the N-layer upsampling network also includes: If the current layer is the first layer: using the convolution module of the current layer to process the target down-sampled feature map output by the encoder in the image segmentation model to obtain a third feature map; The cropping and splicing module is used to crop and splice the down-sampled feature map output by the down-sampling network connected to the current layer and the third feature map to obtain the up-sampled feature map of the current layer.

7. The image processing method according to claim 4, characterized in that: The performing N-layer upsampling processing on the target down-sampled feature map through the N-layer upsampling network also includes: If the current layer is the Nth layer: using the convolution module of the current layer to process the upsampled feature map output by the previous layer to obtain a third feature map; The third feature map is padded with zeros based on the size of the image to be processed to obtain an upsampled feature map of the current layer.

8. An image processing system, characterized in that: include: An acquisition unit, used for acquiring an image to be processed; The encoding unit is used to call the image segmentation model; wherein the encoder in the image segmentation model includes N layers of downsampling networks, each of which includes a convolution module, a first pooling module, a second pooling module, and a dense connection module connected to the first pooling module and the second pooling module; the N layers of downsampling networks in the image segmentation model are used to extract N layers of features from the image to be processed to obtain a target downsampling feature map; wherein N is a non-zero natural number; in each layer of feature extraction process: The convolution module and the first pooling module of the current layer are used in sequence to process the downsampled feature map output by the previous layer to obtain a first feature map; the second pooling module of the current layer is used to perform pooling processing on the image to be processed to obtain a second feature map; the dense connection module of the current layer is used to densely connect the first feature map and the second feature map to obtain a downsampled feature map of the current layer; A decoding unit is used to decode the target down-sampled feature map through a decoder in the image segmentation model to obtain a segmentation result of the image to be processed.

9. A data processing device, characterized in that: include: A processor, wherein the processor is coupled to a memory, wherein the memory stores a program, and wherein the program is executed by the processor, so that the data processing device executes the image processing method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that: The computer storage medium stores computer instructions, and the computer instructions are used to execute the image processing method described in any one of claims 1 to 7.

Images