Detection of high-risk diseases in poultry feces through transfer learning

Abstract

Poultry farming industry constitutes a significant part of the global economy. Deep learning technology possesses the ability to autonomously analyze images, allowing constructed models to aid in the analysis and management of the poultry farming industry, particularly in early detection of sick poultry. Ensuring a sustainable industrial white meat production relies significantly on maintaining a high-quality living environment and early detection of diseases with prompt preventive measures. Early diagnosis of infectious and high-risk diseases such as Coccidiosis, Salmonellosis, and Newcastle disease, coupled with taking necessary precautions, will contribute to the efficient functioning of global economies supply chain. This study aims to detect high-risk Coccidiosis, Salmonellosis, and Newcastle diseases in poultry through transfer learning using poultry feces. Six different transfer learning architectures, namely DenseNet201, Resnet152V2, InceptionV3, InceptionResnetV2, MobileNetV2, and Xception, were employed in the study due to their widespread use and high accuracy rates.

The analysis revealed that MobileNetV2 achieved the highest accuracy rate of 97.1%. Considering the training times, it was observed that MobileNetV2 also exhibited the fastest training. The results of the analysis provide evidence that disease analysis from poultry feces can be carried out with high accuracy through transfer learning in the context of sustainable white meat production.