YOLOPoul: Performance evaluation of a novel YOLO object detectors benchmark for multi-class manure identification to warn about poultry digestive diseases
Digestive diseases are common in poultry and significantly affect their health and productivity. Image processing techniques have gained attention for detecting early signs of disease by analyzing abnormal poultry manure. However, building a reliable system for identifying and locating abnormal manu...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Smart Agricultural Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772375525003776 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Digestive diseases are common in poultry and significantly affect their health and productivity. Image processing techniques have gained attention for detecting early signs of disease by analyzing abnormal poultry manure. However, building a reliable system for identifying and locating abnormal manure in real-world conditions remains challenging and requires large amounts of labeled data for supervised learning. Among deep learning models, the You Only Look Once (YOLO) detector is widely used in precision agriculture due to its speed and accuracy. Herein, a new dataset was created for abnormal poultry manure identification based on updated classification criteria, which included 5688 bounding box annotations across five categories collected from commercial chicken farms. In total, 21 state-of-the-art YOLO models from 8 YOLO versions (YOLOv3–YOLOv9) were fine-tuned and validated to establish comprehensive benchmarks. Detection accuracy in terms of mAP@0.5 ranged from 95.6 % by YOLOv3-tiny to 99.4 % by YOLOv8m, while accuracy in terms of mAP@[0.5:0.95] ranged from 72.9 % by YOLOv4-tiny to 82.2 % by YOLOv9s, with 10 models achieving scores above 80.0 % in mAP@0.5:0.95. High accuracy and efficiency were demonstrated by YOLOv8n and YOLOv8s with inference times under 3 ms. However, traditional data augmentation methods were not fully effective in expanding abnormal manure samples. To address this issue, generative models were explored for data augmentation, where denoising diffusion probabilistic models generated realistic images, showing promising potential. This research provides benchmark data for the classification of abnormal poultry manure, which will become an important resource for promoting future research on poultry disease detection and control based on big data and artificial intelligence, making a fundamental contribution to the field of poultry health monitoring. |
|---|---|
| ISSN: | 2772-3755 |