Applications of Deep Convolutional Neural Network Models for Animal Facial Image Classification : 동물 안면 이미지 분류를 위한 심층 컨볼루션 신경망 모델의 적용

다국어 초록 (Multilingual Abstract)

In first part, the important thing in the field of deep learning is to find out the appropriate hyper-parameter for image classification. In this study, the main objective is to investigate the performance of various hyper-parameters in a convolutional neural network model based on the image classification problem. The dataset was obtained from the Kaggle dataset. The experiment was conducted through different hyper-parameters. For this proposal, Stochastic Gradient Descent without momentum (SGD), Adaptive Moment Estimation (Adam), Adagrad, Adamax optimizer, and the number of batch sizes (16, 32, 64, 120), and the number of epochs (50, 100, 150) were considered as hyper-parameters to determine the losses and accuracy of a model. In addition, Binary Cross-entropy Loss Function (BCLF) was used for evaluating the performance of a model. In this study, the VGG16 convolutional neural network was used for image classification. Empirical results demonstrated that a model had minimum losses obtain by Adagrad optimizer in the case of 16 batch sizes and 50 epochs. In addition, the SGD with a 32 batch sizes and 150 epochs and the Adam with a 64 batch sizes and 50 epochs had the best performance based on the loss value during the training process. Interestingly, the accuracy was higher while performing the Adagrad and Adamax optimizer with a 120 batch sizes and 150 epochs. In this study, the Adagrad optimizer with a 120 batch sizes and 150 epochs performed slightly better among those optimizers. In addition, an increasing number of epochs can improve the performance of accuracy. It can help to create a broader scope for further experiments on several datasets to perceive the suitable hyper-parameters for the convolutional neural network.
In second part, development of livestock has increased demand for identification methods such as deep learning for quality control, welfare management and traceability in a livestock barn. Identification of individual pig has become an issue for traceability in a livestock barn. In this paper, the main objective is to show the feasibility of individual pig face identification and investigate the effects of pig changeable aspects face appearance during growing. Firstly, the datasets were captured in an experimental livestock barn environment at a different time. Secondly, the datasets were filtered similar image by using the structural similarity index measure (SSIM). Thirdly, a face image classification was performed by employing a deep convolutional neural network (DCNN) namely ZFNet model. The results showed that individual pig identification was outperformed while using the same time for training and testing dataset with an accuracy rate above of 97% for each class. The difference between this work and other states of the art pig face recognition work is that training and testing data were captured at 3 different periods in an experimental pig barn.

목차 (Table of Contents)

• Part 1: Analysis the performance of hyper-parameters on deep convolutional neural network of binary classification 3
• I. Introduction 4
• II. Literature Review 6
• III. Materials and Methods 8
• 1. Experiment 8
• 2. Optimizer 8
• 3. Batch Size and Epoch 9
• 4. Loss Function Evaluation Criteria 10
• IV. Results and Discussion 12
• 1. Loss Evaluation 12
• 2. Accuracy Evaluation 16
• 3. Performance of Increasing Batch Size and Epoch 17
• V. Conclusion 20
• Reference 22
• Part 2: Analysis of pig identification of a deep convolution neural network on different age range dataset based on multi classification 25
• I. Introduction 26
• II. Literature Review 28
• III. Materials and Methods 31
• 1. Experiment and Data Collection 31
• 2. Data Filtering 33
• 3. Deep Convolutional Neural Network Architecture Development 35
• 4. Network Hyper-Parameters 36
• 4.1 Optimizer 37
• 4.2 Loss Function 38
• 4.3 Activation Function 38
• 5. System Architecture Specification 39
• 6. Model Evaluation Criteria 39
• IV. Results and Discussion 41
• 1. Accuracy and Loss Performance 41
• 2. Performance of Classification 43
• 2.1 Performance of M1 43
• 2.2 Performance of M2 46
• 2.3 Performance of M3 48
• 2.4 Performance of M4 50
• V. Conclusion 53
• Reference 54