Deep Learning-Based Poultry Health Diagnosis: Detecting Abnormal Feces and Analyzing Vocalizations

This paper introduces a poultry health diagnosis system that employs the You Only Look Once (YOLO) version 7 algorithm and Long Short-Term Memory (LSTM) networks for identifying normal and abnormal chicken fecal images and sounds, including distress calls and illness-related sounds, in poultry houses. Additionally, it utilizes modified generative adversarial networks (GANs) to generate datasets of chicken fecal images. Mel-frequency cepstral coefficients (MFCC) are used for audio sample analysis. Experiment results demonstrate that the system achieves a mean average precision of 86% for diagnosing health based on chicken sounds and 83% for three distinct types of chicken diseases. The system informs users of the recognition results through the LINE notification interface, providing effective warnings. This non-contact detection approach supports the principles of economical animal welfare and reflects a commitment to humane breeding practices.