Title Page
Contents
Abstract 9
1. Introduction 11
2. Review 13
3. Model Improvement 14
3.1. Improvement Methods 14
3.2. Model Structure 20
4. Data and Experimental Methods 25
4.1. Introduction of Datasets 25
4.2. Data Preprocessing 27
4.3. Experimental Methods 28
5. Experimental Results 29
5.1. Experimental Results on CIFAR-10 29
5.2. Experimental Results on CIFAR-100 31
5.3. Experimental Results on DB-MHDI Dataset 33
6. Conclusion 36
7. References 38
Table 1. The highest validation set accuracy achieved during the training process of VGG16, VGG16-MBCNN, Inception10,... 31
Table 2. The highest validation accuracy achieved by Inception10, Inception10-MBCNN, ResNet18, and ResNet18-MBCNN models during... 32
Table 3. Inception10 and Inception10-MBCNN models achieved the highest validation accuracy during the training process on the DB-MHDI dataset. 34
Table 4. ResNet18 and ResNet18-MBCNN models achieved the highest validation accuracy during the training process on the DB-MHDI dataset. 35
Figure 1. Convolutional Neural Network Image Classification Process. 15
Figure 2. Multi-branch Convolutional Neural Network Structure. 16
Figure 3. Convolutional Neural Network with Branches Generated by Number of Classes. 16
Figure 4. Model architectures of VGG16 and VGG16-MBCNN, where N represents the number of samples and K denotes the number of classes. 21
Figure 5. Model architectures of Inception10 and Inception10-MBCNN, where N represents the number of samples and K denotes the number... 22
Figure 6. Model architecture of ResNet18 and ResNet18-MBCNN, where N represents the number of samples and K represents the number of classes. 24
Figure 7. Validation set accuracy during the training process of VGG16, VGG16-MBCNN, Inception10, Inception10-MBCNN, ResNet18, and... 30
Figure 8. Validation Accuracy of Inception10, Inception10-MBCNN, ResNet18, and ResNet18-MBCNN models during training on the... 32
Figure 9. Validation accuracy of Inception10, Inception10-MBCNN, ResNet18, and ResNet18-MBCNN models during training on the... 34