Scopus Indexed Publications

Paper Details


Title
Deep Learning-Based Poultry Disease Classification from Fecal Images: A Comparative Study of CNN Architectures

Author
Md. Ashraf Bin Alam, Abul Hasnat, Hejbulla Asad Shehab, Md. Abir Mahmud Mollah,

Email

Abstract

Poultry production is critical for world food security, but diseases like Newcastle Disease (NCD), Coccidiosis, and Salmonellosis still pose threats to poultry health and productivity. Early diagnosis of diseases with accuracy is highly vital to avoid limiting economic losses and keeping poultry production sustainable. This paper introduces a deep learning-based method for automated poultry disease classification from feces images, offering a non-invasive and streamlined diagnostic tool. A preprocessed and enhanced dataset of 4,090 feces images, split into four classes (NCD, Coccidiosis, Salmonellosis, and healthy), was employed. Five top CNN models, DenseNet169, Tuned DenseNet169, NasNetMobile, MobileNetV2, and Tuned MobileNetV2, were trained using transfer learning with optimization through hyperparameter tuning and cross-validation strategy. While MobileNetV2 and its tuned version show competitive performance (0.95−0.97), DenseNet169 and its tuned counterpart achieve the highest consistency with scores of 0.98 across all trials. Notably, the Tuned DenseNet169 model outperforms others, demonstrating robust and superior accuracy, making it the optimal choice for the given application. The findings demonstrate that deep learning can efficiently classify poultry diseases, offering a scalable and robust means for early disease detection. This research contributes to precision livestock farming via provision of a foundation for non-invasive disease diagnosis that continues to enhance poultry health management as well as reduce economic loss.


Keywords

Journal or Conference Name
2026 International Conference on Artificial Intelligence for Sustainable Engineering and Innovation, AISEI 2026

Publication Year
2026

Indexing
scopus