A Fine Classification Method for Image Objects in Complex Scenes
A complex scene and fine classification technology, applied in the direction of neural learning methods, instruments, biological neural network models, etc., can solve the problems that affect the classification results and cannot automatically focus on the salient regional characteristics of the classification target
Active Publication Date: 2020-04-21
北京同方软件有限公司
View PDF10 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
In addition, in the case of insufficient samples, the self-attention deep learning network cannot automatically focus on the salient regional features of the classification target, and "equal treatment" of salient and non-salient regional features will affect the final classification results
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
[0058] The present invention provides a method for finely classifying image objects in complex scenes, the steps of which are as follows:
[0059] (1) Data preprocessing of image scenes:
[0060] In the complex scene, the detection model is used to clip and select an image of appropriate size 170*170 as the input of the network, and the contrast enhancement operation is performed on the input image data.
[0061] (2) CTreeNet Block feature recalibration:
[0062] see figure 1 , design CTreeNet Block for the feature channel of the network, the method of calculating the attention value of the feature channel is:
[0063] 1) The feature map of the input image after convolution is marked as ,go through convolution
[0064] Feature compression is performed along the spatial dimension, and the convolution is The features are transformed into a size of vector of
[0065] 2) and then vector passing operation to ;
[0066] 3) Features of the input It adopts the ide...
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 a method for finely classifying image targets in complex scenes, which relates to the field of finely classifying images. The method steps of the present invention are: (1) data preprocessing of the image scene; (2) CTreeNet Block feature recalibration; (3) CTreeNet Block and CNN fusion; (4) network structure model establishment; (5) CTreeNet Block Loss calculation in between. Compared with the existing technology, the present invention not only explicitly models the interdependence between feature channels through the CTreeNet network structure, but also adopts the machine learning feature recalibration strategy to automatically obtain each feature channel through the idea of XGBoost algorithm According to this importance, we can improve useful features and suppress features that are not very useful for the current classification task, so as to complete the fine classification of targets in complex scenes.
Description
technical field [0001] The invention relates to the field of fine image classification, in particular to the fine classification of image objects in complex scenes. Background technique [0002] Attention Model (Attention Model) is widely used in various types of deep learning tasks such as natural language processing, image recognition, and speech recognition. It is one of the core technologies worthy of attention and in-depth understanding in deep learning technology. The attention mechanism is a mechanism used in the encoder-decoder structure, and the intuition behind attention can be best explained in terms of human biological systems. In terms of vision, attention depends on how we draw visual attention to different regions of an image in a way that contributes to perception. [0003] From the perspective of the role of Attention, it is divided into two categories: Spatial Attention spatial attention (picture) and Temporal Attention time attention (sequence). For more...
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): G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/213G06F18/24323G06F18/241G06F18/253
Inventor 董小栋赵英郑全新张磊刘阳孟祥松邓家勇江龙赵海波
Owner 北京同方软件有限公司