Implementation Of Convolutional Neural Network With VGG-19 Architecture In Disease Classification In Chickens Based On Faecal Images
Wilis Kaswidjanti, Abdillah Mustamin, Bambang Yuwono, Indah Widowati

UPN Veteran Yogyakarta


Abstract

The objective of this project is to develop an image-based method for diagnosing diseases in poultry using the Convolutional Neural Network (CNN) model with VGG-19 architecture. The chicken is a vital farm animal used to provide eggs, meat, and feathers. But the main problem with raising chickens is that it can be difficult to manually detect diseases by eye inspection. The dataset was improved by applying a range of augmentation techniques, including shift, zoom, and rotation. The data underwent pre-processing to ensure superior image quality. The two main parameters employed in the trials of this research were the number of epochs with a frozen layer and the number of epochs with fine tuning. Thus, the best combination is to fine-tune the entire Convolutional Layer using 20 epochs. Therefore, fine-tuning the complete Convolutional Layer with 20 epochs is the optimal combination. With an accuracy of 97.21%, precision of 97.00%, recall of 97.00%, and f1-score of 97.00%, the final model is rather good. This study significantly advances the accurate and efficient diagnosis of chicken illnesses. Farmers might use the information to determine what preventive actions are necessary.

Keywords: classification, CNN, VGG-19

Topic: Engineering

ICARSESS 2024 Conference | Conference Management System