svnoaie / cat-facia-lpain

cat-facia-lpain

1.For deployment instructions, development environment and system requirements. tensorflow's version is 2.9.1 keras' version is 2.9.0 The spyder is used and the implementation of the system is always executed on a mac system.

2.Introduction to the methodology and documentation of the relevant items. The project is divided into two parts, the first part has three code files, namely 'customfinal.py', 'resnet50.py' and 'Vgg16.py' all three. These models are trained using overall evaluation. The second part is trained on each of the seven features of the cat's face, followed by a decision tree model and a Bayesian model for final pain level recognition. The file mainly consists of a convolutional neural network model with seven different features and a dataset with seven different features with the processing file. A 'main.py' file is included, which introduces the decision tree model and the Bayesian model. The ultimate aim of the project is to classify the level of facial pain in cats by two means: overall recognition classification and feature recognition classification.

3.Technical methods used The architecture used for the system is TensorFlow, Keras.Regarding the framework TensorFlow, it is clear from its name that it consists of two parts: tensor, which stands for tensor, and flow, which stands for the computation of graphs about data flows.TensorFlow is a machine learning system with large-scale learning capabilities.(Martín Abadi. 2016)TensorFlow is supported in a variety of computer languages such as python, java, c++ etc.; and TensorFlow is easy to deploy and can be used on Android as well as ios for relevant recognition and use.It is also partly because Google's use of it has allowed it to become better popular and used.But there is no denying that PyTorch is also a very good deep learning framework. The project is related to machine learning. One of the applications of deep learning is convolutional neural networks. There are three neural network models used in the first approach, namely the custom CNN, ResNet50 and Vgg16. The second method uses mainly convolutional neural networks, decision trees and NaviesBayesian.

4.An introduction to each document. （1）customfinal.py:It introduces about custom CNN models. The internal code includes an introduction to the tensorflow version, the classification of the dataset, the process of building the structure of the custom model, the training panel of the model and the testing panel of the model. （2）resnet 50.py:It presents an introduction to the ResNet50 model. The internal code includes the classification of the dataset, the process of building the structure of the ResNet50 model, the training panel of the model and the testing panel of the model. （3）vgg16.py:It is presented on the Vgg16 model. The internal code includes the classification of the dataset, the process of building the structure of the Vgg16 model, the training panel of the model and the testing panel of the model. （4）main.py:It contains two main models, the decision tree model and the Bayesian model. The code in this file contains mainly a statistical panel for the seven facial features of the cat, a construction panel for the two models, an output panel for the confusion matrix of the two models and an output panel for the evaluation metrics. （5）'feature's label'-preprocessing.py:The main code contained in this file is the data set pre-processing process for each of the different features. （6）‘feature‘s label’cnn.py:This type of file contains the creation of a convolutional neural network on seven features, with the aim of training each feature separately for recognition as well as for classification.