Title Page
Contents
Abstract 10
Part 1. Analysis the Performance of Hyper-Parameters on Deep Convolutional Neural Network of Binary-Classification 12
Ⅰ. Introduction 13
Ⅱ. Literature Review 15
Ⅲ. Materials and Methods 17
1. Experiment 17
2. Optimizer 17
3. Batch Size and Epoch 18
4. Loss Function Evaluation Criteria 19
Ⅳ. Results and Discussion 21
1. Loss Evaluation 21
2. Accuracy Evaluation 25
3. Performance of Increasing Batch Size and Epoch 26
Ⅴ. Conclusion 29
References 31
Part 2. Analysis of Pig Identification of a Deep Convolutional Neural Network on Different Age Range Dataset Based on Multi-Classification 34
Ⅰ. Introduction 35
Ⅱ. Literature Review 37
Ⅲ. Materials and Methods 40
1. Experiment and Data Collection 40
2. Data Filtering 42
3. Deep Convolutional Neural Network Architecture Development 44
4. Network Hyper-Parameters 45
5. System Architecture Specification 48
6. Model Evaluation Criteria 48
Ⅳ. Results and Discussion 50
1. Accuracy and Loss Performance 50
2. Performance of Classification 52
Ⅴ. Conclusion 62
References 63
Part 1: Tables 9
Table 1. Hyper-parameters set up of CNN model for batch size, epoch and optimizer 19
Table 2. The performance of the optimizer hyper-parameter with different number of batch size and epoch by the loss metric 21
Part 2: Tables 9
Table 1. Number of raw images dataset for each pig, training and testing images after applied SSIM. 43
Table 2. Hyper-parameters of the deep convolutional neural network model 46
Table 3. The performance of the deep convolutional neural network to classify the pig face identification accuracy rate of identification. 52
Table 4. The performance matrices of classification model evaluation of Model 1 with testing datasets DS1, DS2, DS3 and DS4 53
Table 5. The performance matrices of classification model evaluation of Model 2 with testing datasets DS1, DS2, DS3 and DS4 55
Table 6. The performance matrices of classification model evaluation of Model 3 with testing datasets DS1, DS2, DS3 and DS4 58
Table 7. The performance matrices of classification model evaluation of Model 4 with testing datasets DS1, DS2, DS3 and DS4 60
Part 1: Figures 7
Fig. 1. Dynamics of loss values of SGD optimizer during training the model through the binary cross-entropy loss function with different Batch Size (BS). (a) 50 Epoch;... 22
Fig. 2. Dynamics of loss values of Adam optimizer during training the model through the binary cross-entropy loss function with different Batch Size (BS). (a)... 23
Fig. 3. Dynamics of loss values of Adagrad optimizer during training the model through the binary cross-entropy loss function with different Batch Size (BS). (a)...[원문불량;p.15] 24
Fig. 4. Represents dynamics of loss values of Adagrad optimizer during training the model through the binary cross-entropy loss function with different Batch Size (BS).... 25
Fig. 5. The training accuracy obtained by SGD optimizer with different number of BS and EP 27
Fig. 6. The training accuracy obtained by Adam optimizer with different number of BS and EP 28
Fig. 7. The training accuracy obtained by Adagrad optimizer with different number of BS and EP 28
Fig. 8. The training accuracy obtained by Adamax optimizer with different number of BS and EP 29
Part 2: Figures 8
Fig. 1. Schematic of pig barn and image of complexity background, noise and light intensity inside a pig barn, Sample of 5 pigs images used in this study. 40
Fig. 2. Data acquisition for collecting pig face image 41
Fig. 3. Camera set up, equipment for collecting pig face image, and sample image from camera 42
Fig. 4. The proposed architecture of convolutional neural network model consisting of five convolutional layers with alternating max pooling, and fully connected layers with softmax... 45
Fig. 5. Training (a) accuracy, (b) loss of training process for each model using convolutional neural network ZFNet based model. 51
Fig. 6. Normalized confusion matrix for multi classification of M1, (a) confusion matrix visualization with testing data DS1; (b) testing data DS2; (c) testing data DS3; (d) testing... 54
Fig. 7. Normalized confusion matrix for multi classification of M2, (a) confusion matrix visualization with testing data DS1; (b) testing data DS2; (c) testing data DS3; (d) testing... 56
Fig. 8. Normalized confusion matrix for multi classification of M3, (a) confusion matrix visualization with testing data DS1; (b) testing data DS2; (c) testing data DS3; (d) testing... 59
Fig. 9. Normalized confusion matrix for multi classification of M4, (a) confusion matrix visualization with testing data DS1; (b) testing data DS2; (c) testing data DS3; (d) testing... 61