An SAR image feature extraction method based on a convolutional neural network
A convolutional neural network and image feature extraction technology, applied in neural learning methods, biological neural network models, neural architectures, etc., can solve problems such as the effect depends on manual experience, the feature extraction effect is unstable, and the algorithm is not universal.
Inactive Publication Date: 2019-06-25
HANGZHOU DIANZI UNIV
View PDF2 Cites 1 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
In actual application, the main problems of the above methods are: human experts are required to design the method of feature extraction and feature selection independently according to specific tasks, and the generality of the algorithm is not strong; the effect of feature extraction depends on manual experience, resulting in the failure of feature extraction Unstable effect
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 moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0018] The present invention is further analyzed below in conjunction with specific examples.
[0019] In this embodiment, the SAR image data of three targets in the MSTAR database are used as the sample data sets for training. In the process of using the SAR image feature extraction method based on convolutional neural network to extract SAR image features, the following steps are specifically included, as follows: figure 1 and figure 2 Shown:
[0020] Step (1), data preprocessing
[0021] Select three types of samples in the MSTAR database, namely T72-132, T62 and 2S1, for each image in the data set, intercept the 128×128 pixel unit containing the target in the image, and add it to the training set.
[0022] Step (2), training convolutional neural network
[0023] Using the training set containing three target categories obtained in step (1), train a convolutional neural network, and the network parameters are shown in Table 1. Stop training when the training error rat...
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
Login to View More Abstract
The invention provides an SAR image feature extraction method based on a convolutional neural network. The method comprises the following steps: firstly, intercepting an area containing a target in anSAR image, and converting the image resolution of an intercepted part into 128 * 128; Training a convolutional neural network by using all converted data; And after the neural network is trained, inputting the SAR image with the features to be extracted into the neural network, and selecting 256-dimensional output of a neural network full connection layer as the extracted features. According to the method, the convolutional neural network is used for feature extraction of the target in the SAR image. The convolutional neural network is used for feature extraction, so that the image preprocessing and feature extraction processes are simplified; Due to the local connection property of convolution and pooling, the translation and rotation of the target in the picture do not affect the finalrecognition effect, so that the network adaptability is stronger.
Description
technical field [0001] The invention belongs to the field of deep learning and image feature extraction, and relates to a Synthetic Aperture Radar (SAR) image feature extraction method based on a convolutional neural network (Convolutional Neural Networks, CNN). Background technique [0002] Synthetic aperture radar has the ability to perform remote sensing monitoring tasks all-weather and all-weather. By coherently accumulating multiple echoes of the target, a two-dimensional SAR image of the target can be obtained. Moreover, microwave remote sensing has a certain ability to penetrate the ground and vegetation, which is helpful to find man-made construction targets such as airports, ports, bridges, and roads, as well as military targets such as aircraft, tanks, and ships. Therefore, it has important military and civil application values. Feature extraction is an important step to obtain effective discriminative features for object recognition in SAR images. The effect of f...
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
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/46G06N3/04G06N3/08
Inventor 薛梦凡韩磊彭冬亮薛安克郭云飞申屠晗骆吉安陈志坤
Owner HANGZHOU DIANZI UNIV