Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Domain adaptive Faster R-CNN (Recurrent Convolutional Neural Network) semi-supervised SAR (Synthetic Aperture Radar) detection method

A semi-supervised and target detection technology, applied in the direction of neural learning methods, instruments, biological neural network models, etc., can solve the problems of test data SAR target detection performance degradation, Faster R-CNN detection performance degradation, etc., to achieve rich data, improve Detection performance, the effect of rich target information

Active Publication Date: 2020-12-22
XIDIAN UNIV
View PDF12 Cites 7 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, Faster R-CNN requires a large amount of labeled training data for network training. When the labeled training data is less, the detection performance of Faster R-CNN drops significantly.
[0005] The existing target detection methods, whether it is traditional CNN or improved Faster R-CNN, are all fully supervised target detection methods. The problem is that a large number of labeled training samples are required for network training, and the labeled training samples In rare cases, these fully supervised target detection methods are prone to overfitting during training, resulting in a significant drop in SAR target detection performance on test data

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Domain adaptive Faster R-CNN (Recurrent Convolutional Neural Network) semi-supervised SAR (Synthetic Aperture Radar) detection method
  • Domain adaptive Faster R-CNN (Recurrent Convolutional Neural Network) semi-supervised SAR (Synthetic Aperture Radar) detection method
  • Domain adaptive Faster R-CNN (Recurrent Convolutional Neural Network) semi-supervised SAR (Synthetic Aperture Radar) detection method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0029] Most of the current SAR target detection methods require a large amount of labeled data for training, and these SAR target detection methods have a strong dependence on labeled samples. Because the number of SAR images is generally small, and marking requires a lot of manpower and material resources, it is difficult to obtain marked SAR images, and usually the data volume of marked SAR images is small. When there are few labeled SAR image training data, the features extracted by these SAR target detection methods have limited expression performance on the target, which further affects the detection performance, resulting in a sharp decline in the detection performance of the current CNN-based SAR target detection methods. Aiming at the above-mentioned status quo, the present invention conducts research and experiments, and proposes a semi-supervised SAR target detection method based on domain-adaptive Faster R-CNN, which can significantly improve the performance of SAR t...

Embodiment 2

[0050] The semi-supervised SAR target detection method based on domain adaptation Faster R-CNN is the same as embodiment 1, in the construction domain adaptation Faster R-CNN model described in the step (3) of the present invention, the basic network of the source domain and the basis of the target domain A constraint item based on the maximum mean difference (MMD) is added between the feature maps extracted by the networks to form an additional domain adaptation module. The expression of the constraint item based on the maximum mean difference (MMD) is as follows:

[0051] The data in the source domain and the target domain obtain corresponding feature maps through their respective basic networks, and the maximum mean difference (MMD) constraint is used between the feature maps of the source domain and target domain data, so that the i-th sample in the source domain and the kth sample of the target domain The corresponding feature map is expressed as and Then the featur...

Embodiment 3

[0057] The semi-supervised SAR target detection method based on domain adaptation Faster R-CNN is the same as embodiment 1-2, in the domain adaptation Faster R-CNN model constructed as described in step (3), the decoder structure connected after the basic network of the target domain It constitutes an additional decoder module, which is used to reconstruct the target domain data. The reconstruction is expressed as a constraint item, as follows:

[0058] The target domain data is adapted to the target domain basic network in Faster R-CNN to extract the feature map of the target domain data, and then the feature map of the target domain data is input into the decoder module, and the output of the decoder module is the input target domain data The reconstruction of , let the decoder for the k-th sample in the target domain data is refactored to Then the reconstructed constraints L recon Expressed as:

[0059]

[0060] Where||·|| 2 Indicates the L2 norm, N t Indicates th...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a semi-supervised SAR (Synthetic Aperture Radar) detection method based on domain-adaptive Faster R-CNN (Recurrent Convolutional Neural Network), which solves the problem thatthe SAR target detection performance is reduced under a small number of marked images. The method comprises the following steps: obtaining a source domain containing a label and target domain data ofa small number of labels; training an original Faster R-CNN by using the source domain data; constructing a domain adaptive Faster R-CNN, initializing the domain adaptive Faster R-CNN, and performingtraining by utilizing source domain and target domain data to obtain a trained model; and inputting the target domain test data into the trained model to obtain a detection result of the test data. According to the method, the domain adaptation Faster R-CNN is constructed, the domain adaptation and decoder module is additionally arranged, SAR target detection is assisted by the optical remote sensing image, dependence on the SAR image with the label is reduced, global information of target domain data is learned through the decoder module, and the detection performance is further improved. Themethod is applied to SAR image target detection.

Description

technical field [0001] The invention belongs to the technical field of radar image processing, and further relates to synthetic aperture radar (SAR, Synthetic Aperture Radar) image automatic target recognition, specifically a semi-supervised SAR target detection method based on domain adaptation Faster R-CNN, which can be used for SAR images Object detection, including vehicle detection for parking lots. Background technique [0002] Synthetic Aperture Radar (SAR) is an active microwave imaging radar, which can realize all-weather and all-weather real-time long-distance monitoring of stationary targets (such as ships, vehicles, etc.). With the rapid development of radar imaging technology, the field of SAR Automatic Target Recognition (ATR) is developing rapidly. Generally, SAR ATR includes the following three steps: target detection, target discrimination and target recognition. As the first step in the SAR ATR steps, target detection is mainly used to determine the posit...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/00G06N3/04G06N3/08
CPCG06N3/08G06V20/13G06V2201/07G06N3/045
Inventor 杜兰廖磊瑶陈健
Owner XIDIAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products