CN-121999361-A - Remote sensing image change detection method and system based on deep learning

CN121999361ACN 121999361 ACN121999361 ACN 121999361ACN-121999361-A

Abstract

The invention provides a remote sensing image change detection method and system based on deep learning. According to the method, a trained remote sensing image change detection model adopts an improved UNet++ network to respectively extract features of an input remote sensing image to be detected, outputs of each layer in the improved UNet++ network are sequentially fused, a designed GC4 module is fused into a feature extraction part of the UNet++ network, the last layer of the feature extraction part and the first layer of an up-sampling part in the UNet++ network are deleted, the number of convolution kernels of all convolution layers in the whole UNet++ network is halved, the improved UNet++ network is obtained, the designed GC4 module replaces a convolution layer in a Bottleneck structure of a C3 module with a Ghost module, a Ghost module is additionally arranged, and meanwhile, a CBAM layer is added between a Concat layer of the C3 module and a Conv layer at the bottom.

Inventors

LI PENGFEI
LI YONGCAI
SUN XIAOPAN
WANG QIAN
ZHAO MINGJUN

Assignees

郑州信大先进技术研究院

Dates

Publication Date: 20260508
Application Date: 20251204
Priority Date: 20251016

Claims (7)

1. The remote sensing image change detection method based on deep learning is characterized by comprising the following steps of: Acquiring remote sensing images to be detected acquired by a target building scene at two different time points; Constructing and training a remote sensing image change detection model to obtain a trained remote sensing image change detection model; The remote sensing image change detection model adopts an improved UNet++ network to respectively extract characteristics of an input remote sensing image to be detected, and sequentially fuses the output of each layer in the improved UNet++ network; The designed GC4 module is fused into a feature extraction part of the UNet++ network, the last layer of the feature extraction part and the first layer of the up-sampling part in the UNet++ network are deleted, and the number of convolution kernels of all convolution layers in the whole UNet++ network is halved at the same time, so that an improved UNet++ network is obtained; the designed GC4 module is characterized in that a convolution layer in a Bottleneck structure of the C3 module is replaced by a Ghost module, and a Ghost module is additionally arranged, meanwhile, a CBAM layer is added between a Concat layer and a Conv layer at the bottom of the C3 module; and inputting the acquired remote sensing image to be detected into a trained remote sensing image change detection model to detect whether the remote sensing image to be detected acquired by the target building scene at two different time points changes or not.
2. The method for detecting the change of the remote sensing image based on the deep learning according to claim 1, wherein a binary cross entropy loss function is selected as a loss function for model training in the process of training the remote sensing image change detection model.
3. A remote sensing image change detection system based on deep learning, comprising: The acquisition module is used for acquiring remote sensing images to be detected acquired by the target building scene at two different time points; The model construction and training module is used for constructing a remote sensing image change detection model and training to obtain a trained remote sensing image change detection model; The remote sensing image change detection model adopts an improved UNet++ network to respectively extract characteristics of an input remote sensing image to be detected, and sequentially fuses the output of each layer in the improved UNet++ network; The designed GC4 module is fused into a feature extraction part of the UNet++ network, the last layer of the feature extraction part and the first layer of the up-sampling part in the UNet++ network are deleted, and the number of convolution kernels of all convolution layers in the whole UNet++ network is halved at the same time, so that an improved UNet++ network is obtained; the designed GC4 module is characterized in that a convolution layer in a Bottleneck structure of the C3 module is replaced by a Ghost module, and a Ghost module is additionally arranged, meanwhile, a CBAM layer is added between a Concat layer and a Conv layer at the bottom of the C3 module; The detection module is used for inputting the acquired remote sensing image to be detected into a trained remote sensing image change detection model so as to detect whether the remote sensing image to be detected acquired by the target building scene at two different time points changes or not.
4. The deep learning-based remote sensing image change detection system according to claim 3, wherein a binary cross entropy loss function is selected as a model training loss function in the process of training the remote sensing image change detection model.
5. An electronic device, comprising: At least one processor, and a memory coupled to the at least one processor; the memory stores a computer program executable by the at least one processor to implement the deep learning-based remote sensing image change detection method according to any one of claims 1 to 2.
6. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed, the method for detecting a change in a remote sensing image based on deep learning according to any one of claims 1 to 2 can be implemented.
7. A computer program product comprising computer program/instructions which, when executed by a processor, implement the method for detecting a change in a remote sensing image based on deep learning as claimed in any one of claims 1 to 2.

Description

Remote sensing image change detection method and system based on deep learning Technical Field The invention belongs to the technical field of image processing, and particularly relates to a remote sensing image change detection method and system based on deep learning. Background The building construction, disassembly, modification, expansion and other change information of the building is very important in the aspects of city planning, city management, illegal building identification and the like. At present, two methods are adopted for statistics of building changes: one method is to examine evidence collection by manual in-situ, the method is difficult to monitor in all directions by only using staff of the natural resource bureau to examine large cities, and the method consumes a great deal of manpower, material resources and financial resources, and has the problems of real-time examination, easy error and the like; the other method is to install high-definition monitoring in the city and make a set of supervision system according to the video detection technology, but the method has higher construction cost, long period and larger limitation, and is difficult to be unfolded and applied in the whole large city. With the deep learning research of people becoming deeper, the deep learning is widely applied to building extraction of remote sensing images, and is used for digital cities, land investigation, military reconnaissance, disaster assessment and the like. For example, a twin network and the like can be used for carrying out depth analysis on pictures before and after the change, and the increased area of the building can be intelligently identified. At present, the building change extraction method has a pixel-based change detection method and a deep learning-based change detection method, and the traditional pixel-based change detection algorithm is easy to generate a salt-and-pepper phenomenon, is easy to be influenced by factors such as illumination and seasonal change, and has the problems of low recognition rate, poor generalization performance and the like. The remote sensing image change detection algorithm based on deep learning, such as a twin network, has high calculation complexity and poor real-time performance, is difficult to be applied to a real-time detection system, and still needs to be further improved in recognition accuracy. Disclosure of Invention The invention aims at overcoming the defects of the prior art, and provides a remote sensing image change detection method and system based on deep learning. Specifically, the first aspect of the present invention provides a remote sensing image change detection method based on deep learning, which includes: Acquiring remote sensing images to be detected acquired by a target building scene at two different time points; Constructing and training a remote sensing image change detection model to obtain a trained remote sensing image change detection model; The remote sensing image change detection model adopts an improved UNet++ network to respectively extract characteristics of an input remote sensing image to be detected, and sequentially fuses the output of each layer in the improved UNet++ network; The designed GC4 module is fused into a feature extraction part of the UNet++ network, the last layer of the feature extraction part and the first layer of the up-sampling part in the UNet++ network are deleted, and the number of convolution kernels of all convolution layers in the whole UNet++ network is halved at the same time, so that an improved UNet++ network is obtained; the designed GC4 module is characterized in that a convolution layer in a Bottleneck structure of the C3 module is replaced by a Ghost module, and a Ghost module is additionally arranged, meanwhile, a CBAM layer is added between a Concat layer and a Conv layer at the bottom of the C3 module; and inputting the acquired remote sensing image to be detected into a trained remote sensing image change detection model to detect whether the remote sensing image to be detected acquired by the target building scene at two different time points changes or not. According to the invention, the GC4 module is used, so that the parameter optimization capability of the Ghost module can be fully utilized, the diversity and adaptability of feature extraction are enhanced, and the capability of the remote sensing image change detection model for remote sensing image change detection is improved. By adding the spatial channel attention mechanism CBAM module after Concat in the GC4 module, the unet++ network can pay attention to important features better, and the performance and feature representation capability of the network are improved, so that feature graphs are adaptively refined to enhance and interactively fuse feature representations, and the distinguishing capability of a remote sensing image change detection model on targets of different change areas is improved. And by del