CN-121999225-A - Image segmentation method, device, medium and product

CN121999225ACN 121999225 ACN121999225 ACN 121999225ACN-121999225-A

Abstract

The invention discloses an image segmentation method, device, medium and product, wherein the method comprises the steps of obtaining a first image, inputting the first image into a first image segmentation model to obtain a first image segmentation result, wherein the first image segmentation model comprises symmetrically arranged multi-level encoders and decoders, each level of the encoder comprises at least one multi-scale depth separable module, each level of the decoder comprises a two-way feature fusion module, and the encoders and the decoders are connected in a cross-level manner through a jump connection module based on a transducer. The technical scheme of the embodiment of the invention can obviously reduce the parameter quantity in the running process of the model, reduce the calculation complexity, maintain the high-precision image processing performance of the model and obtain the high-quality image processing result.

Inventors

HOU WENQIANG
LONG XIAOJING
Niu Donghao

Assignees

中国科学院深圳先进技术研究院

Dates

Publication Date: 20260508
Application Date: 20260130

Claims (10)

1. An image segmentation method, comprising: Acquiring a first image; inputting the first image into a first image segmentation model to obtain a first image segmentation result; The first image segmentation model comprises a multi-level encoder and a decoder which are symmetrically arranged, each level of the encoder comprises at least one multi-scale depth separable module, each level of the decoder comprises a two-way feature fusion module, and the encoder and the decoder are connected in a cross-level mode through a jump connection module based on a transducer.
2. The method of claim 1, wherein the inputting the first image into the first image segmentation model results in a first image segmentation result, comprising: Inputting the first image into a first image segmentation model, and extracting features of the first image for the first time through an input layer of the first image segmentation model to obtain a first feature map; inputting the first feature map to the encoder of the first image segmentation model for second feature extraction to obtain a second feature map; inputting the second feature map to the decoder of the first image segmentation model for feature decoding to obtain a third feature map; And inputting the third feature map to an output layer of the first image segmentation model to carry out segmentation class mapping, so as to obtain the first image segmentation result.
3. The method according to claim 2, wherein the performing, by the input layer of the first image segmentation model, the first feature extraction on the first image to obtain a first feature map includes: Carrying out convolution processing on the first image through a multichannel convolution kernel of the input layer to obtain a multichannel initial feature map; and processing the multi-channel initial feature map through a batch normalization layer and an activation function in the input layer to obtain the first feature map.
4. The method of claim 2, wherein the inputting the first feature map into the encoder of the first image segmentation model performs a second feature extraction to obtain a second feature map, comprising: Inputting the first feature map to the encoder of the first image segmentation model, and extracting the features of the first feature map through a multi-scale depth separable module of a first level of the encoder to obtain an intermediate feature map; sequentially extracting the features in the intermediate feature map through multi-scale depth separable modules of other levels except the first level to obtain a second feature map; Wherein each of the other levels comprises two multi-scale depth separable modules in series.
5. The method according to claim 2, wherein the inputting the second feature map to the decoder of the first image segmentation model performs feature decoding to obtain a third feature map, comprising: Inputting the second feature map to the decoder of the first image segmentation model, and upsampling the input feature map through each level of the decoder to obtain an upsampled feature map; Inputting the up-sampling feature map and the intermediate feature map from the corresponding level of the encoder, which is obtained by the jump connection module, to a two-way feature fusion module to obtain a fusion feature map; And inputting the fusion characteristic map into two serially connected multi-scale depth separable modules to obtain the third characteristic map.
6. The method of any of claims 1-5, wherein the multi-scale depth separable module comprises a depth convolution sub-module and a point convolution sub-module; the depth convolution submodule comprises three-dimensional convolution kernels with the same number as the number of channels of the input feature map, the point convolution submodule comprises a first number of three-dimensional convolution kernels with the size of 1, and each three-dimensional convolution kernel with the size of 1 is used for linearly combining multi-channel features corresponding to the number of channels.
7. The method of claim 6, wherein the processing of the input feature map by the multi-scale depth separable module comprises: And enabling the feature map input into the multi-scale depth separable module to sequentially pass through the depth convolution sub-module, the first batch normalization layer, the first activation function layer, the point convolution sub-module, the second batch normalization layer and the second activation function layer.
8. An electronic device, the electronic device comprising: One or more processors; Storage means for storing one or more programs, The one or more programs, when executed by the one or more processors, cause the one or more processors to implement the image segmentation method as set forth in any one of claims 1-7.
9. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the image segmentation method as claimed in any one of claims 1-7.
10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the image segmentation method as claimed in any one of claims 1-7.

Description

Image segmentation method, device, medium and product Technical Field The present invention relates to the field of medical image processing technologies, and in particular, to an image segmentation method, apparatus, device, medium, and product. Background In the field of 3D medical image segmentation based on deep learning, the three-dimensional medical image segmentation model has huge parameter scale and high computational complexity, so that efficient real-time deployment is difficult to realize in clinical terminal equipment with limited hardware resources. That is, the current method is used for the parameter redundancy of the three-dimensional medical image segmentation model, has high calculation complexity, is seriously dependent on calculation resources, and cannot achieve a balanced state between calculation cost and segmentation quality. Disclosure of Invention The embodiment of the invention provides an image segmentation method, device, medium and product, which can obviously reduce the parameter quantity in the running process of a model, reduce the calculation complexity, maintain the high-precision image processing performance of the model and obtain a high-quality image processing result. In a first aspect, an embodiment of the present invention provides an image segmentation method, including: Acquiring a first image; inputting the first image into a first image segmentation model to obtain a first image segmentation result; The first image segmentation model comprises a symmetrically arranged multi-level encoder and a decoder, each level of the encoder comprises at least one multi-scale depth separable module, each level of the decoder comprises a two-way feature fusion module, and the encoder and the decoder are connected in a cross-level manner through a jump connection module based on a transducer. In a second aspect, an embodiment of the present invention provides an image segmentation apparatus, including: The image acquisition module is used for acquiring a first image; the image segmentation module is used for inputting the first image into the first image segmentation model to obtain a first image segmentation result; The first image segmentation model comprises a symmetrically arranged multi-level encoder and a decoder, each level of the encoder comprises at least one multi-scale depth separable module, each level of the decoder comprises a two-way feature fusion module, and the encoder and the decoder are connected in a cross-level manner through a jump connection module based on a transducer. In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the image segmentation method according to any one of the embodiments of the present invention when executing the program. In a fourth aspect, embodiments of the present invention further provide a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements an image segmentation method according to any of the embodiments of the present invention. In a fifth aspect, embodiments of the present invention also provide a computer program product comprising a computer program which, when executed by a processor, implements the image segmentation method according to any of the embodiments of the present invention. In the embodiment of the invention, a first image is acquired, the first image is input into a first image segmentation model to obtain a first image segmentation result, wherein the first image segmentation model comprises symmetrically arranged multi-level encoders and decoders, each level of the encoders comprises at least one multi-scale depth separable module, each level of the decoders comprises a two-way feature fusion module, and the encoders and the decoders are connected in a cross-level manner through a jump connection module based on a transducer. The technical scheme of the embodiment solves the problems that the calculation complexity is high when the three-dimensional medical image is processed, the accuracy of the image processing result can not be maintained when the calculation complexity is reduced, the parameter quantity in the running process of the model can be obviously reduced, the calculation complexity is reduced, the high-accuracy image processing performance of the model is maintained, and the high-quality image processing result is obtained. Drawings In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to t