An Ensemble Deep Learning Approach to Detect Common Chicken Diseases from Fecal Matter Images

Disease detection from chicken fecal matter images can be a significant step towards automation in the poultry industry.Poultry diseases pose substantial threats to global food security and economic stability.Traditional diagnostic methods are often time-consuming and require specialized skills.In Bangladesh, the poultry sector plays a vital role in fulfilling protein requirements yet faces challenges due to prevalent diseases such as salmonellosis, Newcastle disease, and coccidiosis.Recent advancements in deep learning offer promising alternatives for rapid and non-invasive disease detection.This paper investigates the efficacy of various deep-learning architectures for classifying poultry health based on fecal image analysis.Five prominent models were evaluated: EfficientNetB7, VGG19, Mo-bileNetV3, Vision Transformer, and Swin Transformer, utilizing a publicly available poultry disease dataset from Kaggle.The results demonstrate that EfficientNetB7 achieved 95%, MobileNetV3 91%, VGG19 96%, Swin Transformer 99%, and Vision Transformer 99% accuracy, respectively.The aim is to contribute to the development of a reliable and efficient deep learning-based system for early detection of poultry diseases using fecal samples.Notably, we combined all the models and performed ensemble learning majority voting, achieving a remarkable accuracy of 99.25%.This study underscores the potential of advanced technologies in enhancing disease surveillance and control, offering promising prospects for sustainable and safe poultry farming practices.