Handcrafted feature-based poultry disease detection from fecal images

Poultry farming is the most significant source of the global protein supply chain, which remains highly prone to infectious diseases such as coccidiosis, salmonellosis, and Newcastle disease. This study proposes a handcrafted-based machine learning approach to detect these diseases by analyzing poultry fecal images. We applied multi-color space feature extraction (RGB, HSV, LAB). We explored 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 reduction in dimensionality using RF 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 97.04% accuracy with RF feature selector while requiring no GPU and only 18 minutes of execution time in Google Colab. Compared to deep learning models such as Xception and MobileNetV3, our proposed model offers comparable accuracy with significantly lower usage of computer hardware resources. This work demonstrates a cost-effective, interpretable and scalable alternative to deep learning for the detection of poultry disease in real-time in low-resource agricultural settings. • Lightweight model for poultry disease detection using fecal images. • Multi-color space features (RGB, HSV, LAB) enhance classification accuracy. • Global feature set selected via ablation and feature selection techniques. • ANN classifier achieved 97.04% accuracy without GPU support. • Suitable for low-resource deployment in real-time poultry farms.