A Scalable Deep Learning Framework for Early Detection of Poultry Diseases Using Faecal Imagery

A major sector of agriculture, chicken farming delivers world-wide essential proteins. However, infections include Newcastle disease, Salmonella, and Coccidiosis gravely affect chicken health and economic stability. Conventional diagnosis methods can call for time-consuming laboratory procedures postponing fast treatments. This article suggests a deep learning-based automated and efficient solution based on faecal images for poultry disease identification. We examined a collection of 6,812 faecal images from four classes—Coccidiosis, Salmonella, Newcastle disease, and healthy poultry—which contained 1,361 for validation and 5,451 images for training. A pre-trained MobileNetV2 model was tuned for classification using hyperparameters including a learning rate of 1e-5, batch size of 64, dropout rate of 0.2, and the Adam optimizer. The model derived 94% accuracy with F1 score, precision, and recall values of 0.944, 0.943, and 0.944, respectively. Training made use of early stopping; the model converged in seven epochs, therefore demonstrating efficient learning. The confusion matrix investigation revealed high accuracy over all classes with a modest misclassification bias against minority classes. The results of this work emphasize the possibilities of deep learning in automating illness diagnosis in poultry farming since they present a scalable, non-invasive, fast solution. Using this approach will help vets and farmers to provide accurate and fast diagnosis, hence lowering death rates and improving poultry health control. This study emphasizes the transforming power of artificial intelligence in agriculture, therefore opening the path for sophisticated instruments in environmentally friendly farming methods.