Automated Detection of Poultry Diseases using Implicit Pattern Mining over Explicit Image Analysis

Early detection of poultry diseases is crucial for poultry farmers, but traditional identification methods are often slow and tedious due to limited facilities. We present a comprehensive evaluation of implicit pattern mining techniques using lightweight, CNN-based deep learning models such as, MobileNetV2, InceptionV3, and EfficientNetB5 that are amenable for implementation on edge devices. We contrast them with traditional machine learning approaches applied to analyse chicken fecal image data to classify images of healthy samples from unhealthy ones. We use model interpretation to gain insight to the patterns in the input image that the deep learning models use to make predictions. This facilitates corroborating domain understanding with model performance. This study demonstrates the viability of deploying deep learning based methods for poultry disease detection on handheld devices, highlighting its potential for deployment in the field.