Abstract:Facial recognition using deep learning has been widely used in social life for applications such as authentication, smart door locks, and photo grouping, etc. More and more networks have been developed to facilitate computer vision tasks, such as ResNet, DenseNet, EfficientNet, ConvNeXt, and Siamese networks. However, few studies have systematically compared the advantages and disadvantages of such neural networks in identifying individuals from images, especially for pet animals like cats. In the present study, by systematically comparing the efficacy of different neural networks in cat recognition, we found traditional CNNs trained with transfer learning have better performance than models trained with the fine-tuning method or Siamese networks in individual cat recognition. In addition, ConvNeXt and DenseNet yield significant results which could be further optimized for individual cat recognition in pet stores and in the wild. These results provide a method to improve cat management in pet stores and monitoring of cats in the wild.
What problem does this paper attempt to address?
### Problems the paper attempts to solve
This paper aims to systematically compare the effectiveness of different neural networks in the task of cat individual recognition. Specifically, the researchers focus on the following aspects:
1. **Selection and comparison of neural networks**:
- The researchers selected several commonly - used convolutional neural networks (CNN), including ResNet, DenseNet, EfficientNet and ConvNeXt, as well as the Siamese network.
- By comparing the performance of these networks in the cat individual recognition task, the researchers hope to find the network architecture most suitable for this task.
2. **Comparison of training methods**:
- The researchers used two training methods: fine - tuning and transfer learning.
- By comparing the effects of these two methods, the researchers hope to determine which method is more effective in the cat individual recognition task.
3. **Dataset processing and augmentation**:
- The researchers used a dataset containing 13,536 pictures and pre - processed these pictures, including using the YOLOv5 model to extract the cat's body and face areas, as well as cropping and scaling the images.
- To increase the diversity of the dataset, the researchers also applied a variety of data augmentation techniques, such as flipping, rotating, translating, Gaussian blurring, color jittering, etc.
4. **Evaluation of model performance**:
- The researchers evaluated the performance of different models through the accuracy of the training set, validation set and test set.
- Special attention was paid to the performance of the model on the validation set and test set to evaluate its generalization ability.
### Main findings
- **Performance of traditional CNNs**:
- Traditional CNNs perform well when using the transfer learning method, especially on DenseNet and ConvNeXt.
- The transfer learning method achieved higher accuracy on both the validation set and the test set, with a significant improvement compared to the fine - tuning method.
- **Performance of the Siamese network**:
- Although the Siamese network is generally considered to perform well in face recognition tasks, in the cat individual recognition task, its performance is not as good as that of traditional CNNs.
- The Siamese network shows a certain trend of improvement on the validation set, but its performance on the test set is poor and may need further optimization.
### Conclusion
The study found that traditional CNNs trained with the transfer learning method perform best in the cat individual recognition task, especially DenseNet and ConvNeXt. These results provide new methods for improving the management of pet store and wild cats. However, the performance of the Siamese network in this task is relatively poor and may require more research and optimization.