Wenjun Chen, Chao Cong, Liwen Huang. Convolutional Neural Network Image Feature Measurement Based on Information Entropy[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2211004
- Laser & Optoelectronics Progress
- Vol. 58, Issue 22, 2211004 (2021)
Fig. 1. Examples of different activations of feature layers. (a) Example of fully activated neurons; (b) example of insufficient activation of neurons; (c) example of full activation of deep layer neurons
Fig. 2. Flow chart of image feature measurement algorithm based on information entropy
Fig. 3. Feature purity of each feature layer of ResNet18 and VGG19 under different thresholds. (a) ResNet18; (b) VGG19
Fig. 4. Comparison of Grad-CAM and feature purity of ResNet18 feature layer of different types of images in CIFAR10 dataset
Fig. 5. Comparison of feature activation maps and feature purity of feature layers on ImageNet1000 for models with different performance
Fig. 6. Comparison of Grad-CAM and feature purity of last feature layer under different epochs of Tiny VGG
Fig. 7. Comparison of Grad-CAM and feature purity of last feature layer of ResNet18 model under different epochs
|
Table 1. Comparison of feature purity of different feature layers of VGG and ResNet models
|
Table 2. Comparison of cross-model feature purity on ImageNet1000 dataset
|
Table 3. Comparison of feature purity scores of ResNet18 model and Tiny VGG model under different training epochs
Set citation alerts for the article
Please enter your email address
© Copyright 2018-2021 | Chinese Laser Press. All Rights Reserved 沪ICP备15018463号-20