The Use of Deep Learning Technique in the Classification of Pradu Hang Dam Thai Native Chicken Images สุจิตรา ทิพย์ศรีราช สจี กัณหาเรียง และ สุรชัย สุวรรณลี

Abstract

Once talk about Thai native chicken, Pradu Hang Dam is the most domesticated breed native chicken. However, it has great genetic diversity. Thus, people who has no experience and expertise hardly to distinguish between purebreds and cross breed. In this experiment, designed to use deep learning techniques with convolutional neural networks is presented as an effective image classification technique to solve this problem. Four groups of Pradu Hang Dam Thai native chicken and crossbreed between Pradu Hang Dam and Leung Hang Khao were used as follows: purebred male, purebred female, crossbreed male, and crossbreed female. Image dataset collected 250 images per group, a total of 1,000 images. Four architectures were tested: LeNet-5, AlexNet, CNN1 and CNN2 with resize to 224x224 pixels and then trained at 10 and 20 epochs, respectively. From the experiment it was found that using the LeNet-5 architecture with ReLu and then trained at 20 epochs has the highest training, validation, and testing accuracy but the CNN2 architecture and then trained at 20 epochs able to predict the results with 96.67% accuracy. The results showed that the using a simple architecture of convolutional neural network like CNN2 can efficiently classify Thai native chicken breeds