Poultry farming is a vital component of the global food supply chain, yet it remains highly vulnerable to infectious diseases such as coccidiosis, salmonellosis, and Newcastle disease. This study proposes a lightweight machine learning-based approach to detect these diseases by analyzing poultry fecal images. We utilize multi-color space feature extraction (RGB, HSV, LAB) and explore a wide range of color, texture, and shape-based descriptors, including color histograms, local binary patterns (LBP), wavelet transforms, and edge detectors. Through a systematic ablation study and dimensionality reduction using PCA and XGBoost feature selection, we identify a compact global feature set that balances accuracy and computational efficiency. An artificial neural network (ANN) classifier trained on these features achieved 95.85% accuracy while requiring no GPU and only 638 seconds of execution time in Google Colab. Compared to deep learning models such as Xception and MobileNetV3, our proposed model offers comparable accuracy with drastically lower resource usage. This work demonstrates a cost-effective, interpretable, and scalable alternative to deep learning for real-time poultry disease detection in low-resource agricultural settings.

Coccidiosis, Salmonella, and Newcastle are the common poultry diseases that curtail poultry production if they are not detected early. In Tanzania, these diseases are not detected early due to limited access to agricultural support services by poultry farmers. Deep learning techniques have the potential for early diagnosis of these poultry diseases. In this study, a deep Convolutional Neural Network (CNN) model was developed to diagnose poultry diseases by classifying healthy and unhealthy fecal images. Unhealthy fecal images may be symptomatic of Coccidiosis, Salmonella, and Newcastle diseases. We collected 1,255 laboratory-labeled fecal images and fecal samples used in Polymerase Chain Reaction diagnostics to annotate the laboratory-labeled fecal images. We took 6,812 poultry fecal photos using an Open Data Kit. Agricultural support experts annotated the farm-labeled fecal images. Then we used a baseline CNN model, VGG16, InceptionV3, MobileNetV2, and Xception models. We trained models using farm and laboratory-labeled fecal images and then fine-tuned them. The test set used farm-labeled images. The test accuracies results without fine-tuning were 83.06% for the baseline CNN, 85.85% for VGG16, 94.79% for InceptionV3, 87.46% for MobileNetV2, and 88.27% for Xception. Finetuning while freezing the batch normalization layer improved model accuracies, resulting in 95.01% for VGG16, 95.45% for InceptionV3, 98.02% for MobileNetV2, and 98.24% for Xception, with F1 scores for all...