A Rotation Invariant Semantic Information Mining Method
A technology of invariant rotation and information blocks, applied in neural learning methods, digital data information retrieval, instruments, etc., can solve problems such as interference, increase overall feature differentiation, and achieve the effect of improving differentiation
Active Publication Date: 2022-04-01
HANGZHOU DIANZI UNIV
View PDF7 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
If at the same time mining and using the semantic information in the upper right corner of the image feature to enhance the overall feature, not only will not increase the distinction of the overall feature, but will introduce interference features
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
[0024] The purpose and effects of the present invention will become more apparent by referring to the accompanying drawings in detail of the present invention.
[0025] The present invention utilizes the ring segmentation strategy ( figure 1 As shown), the rotation-invariant image semantic information is mined to enhance the discriminativeness of image feature descriptors and improve the accuracy of cross-view geographic image retrieval. Its overall flow chart is as follows figure 2 As shown, the specific steps are as follows:
[0026] Step 1: Build a feature generation network. The network consists of three parts: the first part is the ResNet-50 network, which is used to extract the feature map; the second part is the main branch feature processing network, which performs average pooling and dimensionality reduction processing on the feature map; the third part is the auxiliary branch Feature processing network, this part uses a ring segmentation strategy to mine rotation...
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 provides a rotation-invariant image semantic information mining method. First, a feature generation network is constructed, and the ResNet-50 network feature map is extracted; the feature map is averaged pooled and dimensionally reduced through the main branch feature processing network; Mining rotation-invariant image semantic descriptors with side-branch feature processing networks. An n-dimensional feature obtained by the main branch feature processing network and multiple n-dimensional features obtained by the auxiliary branch feature processing network are used for feature splicing to obtain an enhanced image feature descriptor. Finally, the enhanced feature descriptor is used to retrieve images from different perspectives, and then realize geographic target positioning. The invention proposes a ring segmentation strategy, so that the acquired semantic blocks will not be disturbed by the shooting direction, and the mined semantic blocks can improve the distinction of image features.
Description
technical field [0001] The invention relates to the field of image retrieval, in particular to a rotation-invariant image semantic information mining method. Background technique [0002] The task of image retrieval has long been an important research topic in the field of computer vision, and its purpose is to quickly find images that meet the conditions in a huge image database. Image retrieval can be subdivided into different sub-tasks according to different application fields, such as: pedestrian re-identification, vehicle re-identification, cross-view geolocation, etc. The present invention is mainly applied to cross-viewpoint geolocation tasks. [0003] Cross-view geolocation aims to retrieve two images with the same semantic information from different viewpoints. Cross-view geolocation has been widely used in scenarios such as drone precision delivery, robot navigation, and event detection. Taking UAV delivery as an example, given a satellite image with GPS locatio...
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 Patent Type & Authority Patents(China)
IPC IPC(8): G06F16/55G06F16/583G06V10/44G06V10/764G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06F16/583G06F16/55G06N3/08G06V10/44G06N3/045G06F18/241
Inventor 颜成钢王廷宇万斌孙垚棋张继勇张勇东
Owner HANGZHOU DIANZI UNIV