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

Target detection method based on dense connection deep network

A dense connection, target detection technology, applied in neural learning methods, biological neural network models, instruments, etc., can solve problems such as increased network computing, limited network applications, large video memory, etc., to strengthen feature propagation and improve detection results. , the effect of enhanced learning

Active Publication Date: 2020-04-10
JIANGNAN UNIV
View PDF4 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] Densely Connected Convolutional Neural Networks (Gao Huang, Zhuang Li, Laurens van der Maaten, Kilian Q. Weinberger. Densely Connected Convolutional Networks [C]. CVPR, 2017. DOI: 10.1109 / CVPR.2017.243) is an independent and complete detection network, but due to the settings of different convolutional layer output parameters and the existence of the fully connected layer, the network's calculation load has increased sharply, and a large amount of video memory has to be consumed.
This problem limits the application of this network in actual production

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
  • Target detection method based on dense connection deep network
  • Target detection method based on dense connection deep network
  • Target detection method based on dense connection deep network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0052] Although the existing object detection methods have high detection accuracy, they cannot meet the requirements of real-time detection in actual production, and their portability is poor due to the model memory. In view of these problems, the present invention proposes a kind of target detection method based on densely connected deep network, which will be described in detail below in conjunction with the accompanying drawings:

[0053] Such as figure 1 As shown, it is a structural diagram of a densely connected network based on a densely connected deep network target detection method provided by the present invention; as figure 2 As shown, it is a network overall architecture diagram of a densely connected deep network-based object detection method provided by the present invention. In this embodiment, a target detection method based on a densely connected deep network includes the following parts:

[0054] A.1. Read the image data in the Pascal VOC dataset and extra...

Embodiment 2

[0079] This embodiment is the process and results of human detection on the Pascal VOC dataset. Specific steps are as follows:

[0080] A.1 Read in the pedestrian image data in different scenes in the Pascal VOC dataset as training data and extract pedestrian data features: the network reads the input image, first normalizes its resolution to 416*416, and then passes through the convolutional layer , pooling layer and dense connection module (Dense Block) extract and fuse the features of each channel.

[0081] Said step A.1 comprises:

[0082] (1) The dense connection method introduced makes the L-layer network have L(L+1) / 2 connections. The dense connection module Dense Block is mainly composed of 1*1 and 3*3 convolution layers, and the 1*1 convolution operation is also called bottleneck layer, the purpose is to reduce the number of input feature maps, improve computational efficiency and integrate characteristics of each channel. 3*3 convolution is used to extract image ...

Embodiment 3

[0107] This embodiment is the process and result of detecting the category of horses on the Pascal VOC dataset. Specific steps are as follows:

[0108] A.1 Read in the image data of horses in different scenes in the Pascal VOC dataset as training data and extract the characteristics of horse category data: the network reads the input image, first normalizes its resolution to 416*416, and then passes The convolutional layer, pooling layer, and dense connection module (Dense Block) extract and fuse the features of each channel.

[0109] Said step A.1 comprises:

[0110] (1) The dense connection method introduced makes the L-layer network have L(L+1) / 2 connections. The Dense Block is mainly composed of 1*1 and 3*3 convolutional layers, and the 1*1 convolution operation is also called the bottleneck layer. The purpose is to reduce the number of input feature maps, improve computational efficiency and integrate various channels. Characteristics. 3*3 convolution is used to extra...

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 target detection method based on a dense connection deep network, belonging to the technical field of target detection. According to the target detection method based on thedense connection deep network, a dense connection mode is fused into the yolo-tiny network, a convolution layer is increased, and a feature extraction network is improved; and the improved network firstly normalizes an input image into a fixed size, then uses a DenseBlock module to extract and fuse the characteristics of each channel, and then uses different priors anchors for prediction on different scales to complete the classification and positioning of a target. Compared with an original algorithm, the improved algorithm has the advantages that precision is improved by 15%, and the requirement of real-time detection can still be met; and the size of a model is only 44.7 MB, so the requirements on memory occupation and real-time performance in actual use can be met.

Description

technical field [0001] The invention relates to a target detection method based on a densely connected deep network, and belongs to the technical field of target detection. Background technique [0002] There are currently many target detection algorithms based on deep learning, such as Faster Rcnn (Faster Region-based Convolutional Network), SSD (Single Shot MultiBox Detector), R-fcn (Region-based Fully Convolutional Networks), yolo (You Only Look Once) , yolo-tiny (YouOnly Look Once-Tiny) and so on. However, these algorithms still have many shortcomings. For example, Faster Rcnn, R-fcn, SSD and other algorithms have problems such as slow detection speed and complex system configuration environment. The yolov3 algorithm has a faster detection speed, but the model takes up a lot of memory, while yolov3-tiny There is a problem that the detection accuracy is too low. [0003] Although the current yolov3-tiny detection network has a fast detection speed, there are various pro...

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): G06K9/00G06K9/62G06N3/04G06N3/08
CPCG06N3/084G06V20/46G06N3/045G06F18/23213G06F18/253
Inventor 陈莹潘志浩化春键
Owner JIANGNAN 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

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com