Check patentability & draft patents in minutes with Patsnap Eureka AI!

Class activation mapping method

A mapping method and type of technology, which is applied in the field of class activation map generation research, can solve the problems of class activation map quality impact, gradient instability, etc., and achieve the effect of improved accuracy and good quality

Pending Publication Date: 2020-08-18
HARBIN ENG UNIV
View PDF4 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to solve the problem that the quality of the class activation map generated with the gradient as the weight is affected due to the instability of the gradient, and proposes an Acc-CAM class activation mapping method

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
  • Class activation mapping method
  • Class activation mapping method
  • Class activation mapping method

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0023] Specific embodiment one: a kind of Acc-CAM class activation mapping method described in the present embodiment, this method comprises the following steps:

[0024] Step 1, set up AlexNet model, the network structure of described AlexNet model comprises several convolutional layers and several fully connected layers, and is connected with Softmax function behind the last fully connected layer;

[0025] After training the established AlexNet model, obtain the trained AlexNet model;

[0026] Step 2: Input a single original image sample into the trained AlexNet model, extract N original activation maps through the convolutional layer of the AlexNet model, and then restore the extracted N original activation maps to the size of the original image sample , to obtain N size-restored activation maps, where N is the number of convolution kernels in the last convolutional layer of the AlexNet model;

[0027] Step 3, performing normalization processing on the pixels in the image ...

specific Embodiment approach 2

[0041] Specific implementation mode two: combination figure 1 This embodiment will be described. The difference between this embodiment and the first embodiment is that in the first step, the network structure of the AlexNet model has seven layers in total, among which, the first five layers are all convolutional layers, and the last two layers are all fully connected layers.

[0042] The last fully connected layer is connected to the Softmax function, which produces a certain number of distributions of class labels. The specific parameters of the AlexNet model are shown in Table 1.

[0043] Table 1 Model parameter list

[0044]

specific Embodiment approach 3

[0045] Specific embodiment three: the difference between this embodiment and specific embodiment two is that: the N original activation maps extracted in step two are the activation maps obtained through the last convolutional layer;

[0046] Then the extracted N original activation maps are respectively restored to the size of the original image sample, and the deconvolution operation method is used to restore the size of the original activation map.

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 class activation mapping method, and belongs to the technical field of class activation mapping graph generation research. According to the class activation mapping method, the problem that the quality of the class activation mapping graph generated by taking the gradient as the weight is influenced due to instability of the gradient is solved. According to the class activation mapping method, the activation graph obtained by the last convolution layer in the AlexNet model is restored to the size of the input image; a treatment mode similar to a mask is adopted; however, the difference is that the pixel value of the original input image and the corresponding pixel value in the activation image with the same size are subjected to dot multiplication operation, and then the generated mask image samples are input into the AlexNet model, and then the value of each mask image sample is obtained through a Softmax function, and the obtained value of each mask image sample is taken as the weight value of the corresponding activation graph; and then linear weighting is performed on the restored activation graph with the same size and the weight to obtain a final Acc-CAM class activation mapping graph. The class activation mapping method can be applied to generation of the class activation mapping graph.

Description

technical field [0001] The invention belongs to the technical field of class activation map generation research, and in particular relates to an Acc-CAM (Accuracy-Weighted Class Activation Mapping) class activation mapping method. Background technique [0002] Convolutional neural networks have shown tremendous effectiveness in many practical tasks, but when the model misbehaves, it often produces unexplainable and incoherent results, making one wonder what caused the neural network to make such a decision . For ordinary users, using a deep neural network model is like a black box. Give it an input, and it will feed back a decision result. No one can know exactly the decision basis behind it and whether the decision it makes is reliable. Therefore, the lack of interpretability has become one of the main obstacles for the further development and application of deep learning in real-world tasks. In recent years, although network architectures have been continuously simplifie...

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
IPC IPC(8): G06N3/04
CPCG06N3/045
Inventor 王念滨张英琪张耘王红滨周连科张毅厉原通
Owner HARBIN ENG UNIV
Features
  • R&D
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More

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

© 2025 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com