Remote sensing image target detection method based on deep evolution pruning convolutional network

A target detection and depth convolution technology, applied in the field of image processing, can solve the problems affecting the computational complexity and computational speed of the model, difficult optical remote sensing image detection, and overall accuracy loss, so as to overcome the loss of model operation speed and reduce model parameters. amount, the effect of accelerating the convergence speed

Active Publication Date: 2019-12-03
XIDIAN UNIV
View PDF8 Cites 33 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] In the prior art, due to the large size of the remote sensing image, low resolution, small target size and blurred target edge, the existing methods often fail to learn the characteristics of the target when performing remote sensing image target detection, which leads to the failure of target detection. The accuracy rate is low, and due to the huge amount of data in remote sensing images and the huge amount of parameters in the network model, the detection speed is greatly limited
[0004] The efficiency and accuracy of existing target detection technologies are often incompatible
The second-order detection model such as FasterR-CNN has high accuracy, but also brings a huge amount of calculation; the first-order detection model such as YOLO and SSD has a faster calculation speed, but the accuracy rate is not satisfactory
However, the shortcomings of this method are that there is a large amount of redundant information in the residual network ResNet and the feature pyramid

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
  • Remote sensing image target detection method based on deep evolution pruning convolutional network
  • Remote sensing image target detection method based on deep evolution pruning convolutional network
  • Remote sensing image target detection method based on deep evolution pruning convolutional network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0037] Remote sensing image target detection is an application that has attracted much attention in the field of remote sensing image processing and analysis, for example, to determine whether there are targets such as aircraft and ships in remote sensing images, and to identify, classify and accurately locate them. With the continuous development of satellite technology, the amount of existing optical remote sensing image data is becoming larger and larger, and compared with the vast sea area, the target size of aircraft and ships is small and sparse. Detecting objects accurately is a challenging task. The existing remote sensing image target detection technology often focuses on how to better learn the characteristic information of the target, and then improve the accuracy of target detection. However, due to the huge amount of data in remote sensing images and the huge amount of parameters in the network model, the current detection speed Extremely restricted.

[0038] The...

Embodiment 2

[0071] The remote sensing image target detection method based on depth evolutionary pruning convolutional network is the same as embodiment 1, step (3) describes the construction of deep convolutional feature extraction sub-network, and its specific steps are:

[0072] (3a) Build a depth-separable convolution anti-residual connection module: its module structure is, feature map input layer in the previous stage → 1×1 convolution layer → depth-separable convolution unit → point-by-point addition layer → current stage Feature map output layer.

[0073] In the anti-residual connection module, the 1×1 convolutional layer and the depth-separable convolution unit appear in groups, and the point-by-point addition layer is to combine the output feature map of the depth-separable convolution unit of the previous layer with the output feature map from the anti-residual connection The feature map of the module input layer is added point by point to form a feature processing layer.

[0074...

Embodiment 3

[0080] The remote sensing image target detection method based on depth evolutionary pruning convolution network is the same as embodiment 1-2, the depth separable convolution unit described in step (3a), see figure 2 , its unit structure is, feature map input layer in the previous stage → 3×3 deep convolution layer → first batch normalization layer → ReLU activation function layer → 1×1 pointwise convolution layer → second batch normalization layer Normalization layer → linear activation function layer → output feature layer.

[0081] The depthwise separable convolution unit integrates the standard convolution into depthwise convolution and pointwise convolution to realize the space and channel separation of features and separate processing, thereby greatly reducing the amount of parameters and computational complexity.

[0082] The activation function after the 1×1 point-by-point convolution layer no longer uses the ReLU activation function, but uses a linear activation func...

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 remote sensing image target detection method based on a deep evolution pruning convolutional network. The problem that the detection speed and the detection precision are notglobally and effectively optimized at the same time in existing remote sensing image target detection is solved. The method comprises the specific steps of processing a data set; constructing a deepconvolution feature extraction subnet; constructing a full convolution FCN detection subnet; constructing and training a deep convolution target detection network; constructing and training a target detection network based on a deep evolution pruning convolutional network; performing target detection on the test data set by using the trained model; and outputting test results. According to the method, a reverse residual structure is constructed by using depth separable convolution, so that the model parameter quantity is greatly reduced while the detection precision is high; the target detection network is combined with evolutionary pruning to realize global acceleration. The method greatly reduces the calculation amount, remarkably improves the target detection speed, is high in detectionprecision, and is used for the quick and accurate detection of small targets in a remote sensing image, such as an airplane and a ship.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to remote sensing image target detection, specifically a remote sensing image target detection method based on deep evolutionary pruning convolutional network, which can be applied to aircrafts and ships in different areas in remote sensing images ground object detection. Background technique [0002] Target detection technology is one of the core issues in the field of computer vision. Remote sensing image target detection refers to the use of images captured by remote sensing satellites as data sources, and the use of image processing technology to locate and classify objects of interest in the images. Remote sensing image target detection, as a key technology in the application of remote sensing images, can capture attack targets and provide accurate location and category information in high-tech military confrontation. It has a significant impact on the military 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
no application Login to view more
IPC IPC(8): G06K9/00G06K9/62G06N3/04G06N3/08
CPCG06N3/082G06V20/13G06N3/045G06F18/214
Inventor 焦李成李玲玲姜升郭雨薇程曦娜丁静怡张梦璇杨淑媛侯彪
Owner XIDIAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap