A small object detection method in real scene based on multi-task generation antagonistic network

A detection method and real scene technology, applied in the field of computer vision and object detection, can solve the problems of small objects cannot be clearly identified, poor adaptive ability, etc.

Active Publication Date: 2019-01-04
HARBIN INST OF TECH
View PDF6 Cites 11 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problem that the existing small object detection method has poor adaptability to the complex and changeable working environment in the real scene, which leads to the problem that small objects cannot be clearly identified in the case of complex and changeable backgrounds, a method based on multi-task generation is proposed. A small object detection method in a real scene against the network. The small object detection method based on the multi-task generation against the network makes the detection object not only limited to the larger objects in the real scene, but also not limited to the ideal situation in the laboratory. pictures, and realize the detection of small objects in real complex scenes, especially the detection of small objects generated when the object is far away from the image capture device

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 small object detection method in real scene based on multi-task generation antagonistic network
  • A small object detection method in real scene based on multi-task generation antagonistic network
  • A small object detection method in real scene based on multi-task generation antagonistic network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0060] This embodiment proposes a real-scene small object detection method based on a multi-task generative confrontation network, such as Figure 4 As shown, the detection method includes:

[0061] Step 1: determine a training sample set, and use the training sample set to train an object detector; the training sample set contains a plurality of images;

[0062] Step 2: using the object detector to intercept and generate training samples for the multi-task generation confrontation network;

[0063] Step 3: constructing a multi-task generative adversarial network training sample according to the training sample; the multi-task generative adversarial network training sample includes a positive training sample and a negative training sample;

[0064] Step 4: taking the training sample generated by interception as a high-resolution image, and using bilinear interpolation to downsample the high-resolution image by 4 times to obtain an image as a corresponding low-resolution image...

Embodiment 2

[0072] This embodiment is a further limitation of the method for detecting small objects in real scenes described in Embodiment 1. The training sample set can be collected by itself according to actual needs, and then a corresponding object category database can be constructed, or an existing public object detection database can be selected. Such as Pascal VOC, Microsoft COCO and other databases. In order to facilitate comparison with other existing methods, this embodiment uses the Microsoft COCO data set as the training sample set, and uses the 115K data in the Microsoft COCO data set as the training set sample, and uses the 5K data in the Microsoft COCO data set as the verification set sample, Use the 5K data in the Microsoft COCO dataset as a test set sample;

[0073] The object images in the training samples are divided into three grades of large, medium and small according to the size of the area, wherein a larger object image refers to an image with an area larger than ...

Embodiment 3

[0076] This embodiment is a further limitation of the method for detecting small objects in real scenes described in Embodiment 1 or 2. The specific process for the object detector in step 2 to intercept and generate training samples for a multi-task generation adversarial network includes:

[0077] The first step: using the object detector to predict object position information for each image in the training sample set;

[0078] Step 2: Intercept 600 images of areas most likely to contain objects in each image after object position information prediction;

[0079] The third step: saving the intercepted image area, and the saved area image is used as a training sample of the multi-task generation confrontation network.

[0080] In this embodiment, the object detector is used to predict the object position information of each image in the Microsoft COCO training sample set, and 600 regions most likely to contain objects are intercepted from each image and saved, and these saved...

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 provides a real scene small object detection method based on a multi-task generation antagonism network, belonging to the technical field of object detection in the computer vision field. The detection method introduces a multi-task generation antagonistic network into object detection, and realizes the establishment of a small object detection network and the completion of the smallobject detection of an image by making the generator network and the discriminator network in the multi-task generation antagonistic network learn from each other in a game mode. The small object detection method overcomes the difficulty of low detection accuracy of the object detection method in the real scene at the present stage, and promotes the application of the object detection method based on depth learning in the real scene.

Description

technical field [0001] The invention relates to a method for detecting small objects in a real scene based on a multi-task generation confrontation network, and belongs to the technical field of object detection in the field of computer vision. Background technique [0002] In recent years, driven by a new round of "artificial intelligence revolution", the competition in science and technology has become increasingly fierce, and the rapid development of artificial intelligence has profoundly changed human social life and changed the world. The object detection technology based on deep learning, especially the small object detection technology in real scenes, is a solution for specific object recognition and positioning that emerged and developed rapidly under this background. As an important part of the image processing field, object detection technology has always been a very important basic research topic in the field of computer vision. At the same time, it plays a key ro...

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): G06K9/62G06N3/04
CPCG06N3/045G06F18/214
Inventor 张永强丁明理李贤杨光磊董娜朱乐熠白延成
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