A remote sensing image ship target detection method based on sparse mobilenetv2 network

A remote sensing image and target detection technology, which is applied in the field of remote sensing image ship target detection, can solve the problems of redundant remote sensing image information, large amount of remote sensing image data information, and difficulties in real-time analysis of satellites in orbit.

Active Publication Date: 2021-04-20
HARBIN INST OF TECH
View PDF8 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to solve the difficulty in extracting target features due to the redundancy of ship remote sensing image information, and the difficulty in real-time analysis of satellites on orbit due to the contradiction between the large amount of remote sensing image data information and the weak on-orbit calculation and processing capabilities. problem, and proposed a remote sensing image ship target detection method based on sparse MobileNetV2 network

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
  • A remote sensing image ship target detection method based on sparse mobilenetv2 network
  • A remote sensing image ship target detection method based on sparse mobilenetv2 network
  • A remote sensing image ship target detection method based on sparse mobilenetv2 network

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0018] Specific implementation mode one: as figure 1 As shown, a remote sensing image ship target detection method based on a sparse MobileNetV2 network described in this embodiment, the method includes the following steps:

[0019] Step 1. Carry out overlapping cutting on the original remote sensing image according to the overlapping ratio α to obtain the cut image;

[0020] Step 2, downsampling the image block in the cut image in step 1, to obtain the downsampled image block;

[0021] Step 3. Establish a sparse MobileNetV2 network consisting of a seven-level convolutional layer and a pruned MobileNetV2 network;

[0022] And use the down-sampled image blocks in step 2 to train the seven-level convolutional layer, and use the image blocks in the cut image in step 1 to train the pruned MobileNetV2 network. When the number of training times reaches the set maximum number of iterations Stop iterating when , and obtain the trained sparse MobileNetV2 network;

[0023] Step 4. Fo...

specific Embodiment approach 2

[0027] Specific implementation mode two: the difference between this implementation mode and specific implementation mode one is: the specific process of the step one is:

[0028] According to the overlapping ratio α, the original remote sensing image with a size of 16k×16k pixels is overlapped and segmented to obtain the segmented image; in the segmented image, the size of each image block is 512×512 pixels.

[0029] The purpose of this embodiment is to make the ship target account for more than one-third of the small block image where it is located, and the purpose of overlapping is to ensure the integrity of the target and improve the accuracy of detection. Such as Figure 5 Shown is a schematic diagram of overlapping segmentation of the original remote sensing image.

specific Embodiment approach 3

[0030] Embodiment 3: This embodiment is different from Embodiment 2 in that: the image block included in the cut image is down-sampled, and the down-sampling method is a bilinear interpolation method.

[0031] The specific process of downsampling in this embodiment is as follows: the downsampling adopts the existing bilinear interpolation method, which is also called quadratic linear interpolation method. It uses the pixel values ​​of the corresponding four points in the original image to determine the pixel values ​​in the target image.

[0032] pseudocode

[0033] enter:

[0034] Img: original image

[0035] zmf: zoom factor

[0036] output:

[0037] new_img: output image

[0038] Step1: Find the size of the original image Img, which is recorded as height×width×channel, and then generate a matrix new_img with all 0s whose size is (zmf×height)×(zmf×width)×channel;

[0039] Step2: Extend the boundary of Img to get IT, the size is (height+2)×(width+2)×channel;

[0040] S...

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

A remote sensing image ship target detection method based on a sparse MobileNetV2 network belongs to the technical field of remote sensing image ship target detection. The invention solves the problems of difficulty in extracting target features due to redundancy of ship remote sensing image information, and difficulty in real-time analysis of satellites on orbit due to the contradiction between large amount of remote sensing image data information and weak on-orbit calculation and processing capabilities. The present invention performs overlapping segmentation and down-sampling processing on remote sensing images to obtain down-sampled image blocks; uses the down-sampled image blocks to train seven groups of convolutional layers, uses the image blocks in the overlapped and segmented images to train the pruned The MobileNetV2 network is trained, and the trained sparse MobileNetV2 network is obtained when the maximum number of iterations is reached; the trained sparse MobileNetV2 network is used to perform target detection on the remote sensing image to be measured. The invention can be applied to the technical field of remote sensing image ship target detection.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image ship target detection, and in particular relates to a remote sensing image ship target detection method. Background technique [0002] In recent years, with the rapid development of aerospace remote sensing technology, remote sensing image processing technology has been widely used in military and civilian fields, especially target detection technology based on remote sensing images has become an important technical support for military reconnaissance, search and rescue operations. As the main transport carrier and important military target at sea, the detection and identification of ships is helpful for the dispatch and deployment of key sea areas, assisting in the rescue of ships in distress, and combating illegal acts at sea. Remote sensing images provide an extremely rich data source for them. However, the emergence of massive remote sensing data also brings more complex applicati...

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 Patents(China)
IPC IPC(8): G06K9/00
CPCG06V20/13G06V2201/07
Inventor 彭宇尹童马宁于金祥彭喜元
Owner HARBIN INST OF TECH
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