SSD network construction method and target detection method based on BiFPN enhanced feature extraction
A technology of feature extraction and network construction, applied in the direction of neural learning methods, biological neural network models, neural architectures, etc., can solve the problems of poor recognition accuracy and achieve the effects of improving detection accuracy, good recognition effect, and enhancing key information
Pending Publication Date: 2022-05-17
XI'AN UNIVERSITY OF ARCHITECTURE AND TECHNOLOGY
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0006] In order to improve the problem of poor recognition accuracy of difficult targets such as small targets by the original SSD network, the purpose of the present invention is to propose a SSD target detection method based on BiFPN enhanced feature extraction
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
[0042] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0043] Such as figure 1 As shown in the traditional SSD network, the construction idea is as follows:
[0044] The SSD network is a classic single-target detection algorithm. Its advantage is that target detection and classification are completed at the same time. While ensuring accuracy, it also has fast detection performance. The structure principle of SSD network is as follows: figure 1 As shown, the input image size is unified to 300×300, and through the feature extraction network, six effective feature layers are obtained: Conv4_3, fc7, Conv8_2, Conv9_2, Conv10_2, Conv11_2. The SSD network can be divided into two parts: the VGG network (Conv1 to fc7) and four additional convolutional layers (Conv8 to Conv11). The VGG network is the basic backbone network for shallow feature extraction. In the end, only Conv4_3 and fc7 are sent to the ...
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 discloses an SSD network construction method based on BiFPN enhanced feature extraction, and the method comprises the following steps: 1, transmitting six effective feature layers Conv43, fc7, Conv82, Conv92, Conv102 and Conv111 obtained by an SSD network to an enhanced feature extraction network; step 2, performing up-sampling feature fusion on the six input effective feature layers by the enhanced feature extraction network, and then performing down-sampling feature fusion to obtain six new feature layers; and step 3, adding six new feature layers obtained by the enhanced feature extraction network into a channel attention mechanism to obtain an improved SSD network. And step 4, performing network model training on the improved SSD network obtained in the step 3 to obtain a trained SSD network model. The problem that a traditional SSD network is poor in recognition precision of difficult targets such as small targets is effectively solved, and through tests, the average detection precision on a VOC2007 + 2012 data set is 79.4% and is improved by 2.2% compared with traditional SSD.
Description
technical field [0001] The invention belongs to the technical field of target detection based on deep learning, and relates to an SSD network construction method and a target detection method based on BiFPN enhanced feature extraction. Background technique [0002] Object detection technology is an important research content and application field in the field of computer vision. In many scenarios such as pedestrian detection, vehicle recognition, unmanned driving and remote sensing image detection, it has very important research value and significance. People are more and more willing to put target detection technology into engineering practice. Target detection technology is gradually It is closely related to our daily life. [0003] How to further improve and enhance the target detection technology has become the focus and difficulty of current research at home and abroad. The early target detection was mainly a combination of feature extraction and classification. The t...
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 IPC IPC(8): G06V20/00G06V10/764G06V10/80G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06N3/082G06N3/045G06F18/2431G06F18/253
Inventor 马宗方徐静冉罗婵宋琳张国飞
Owner XI'AN UNIVERSITY OF ARCHITECTURE AND TECHNOLOGY