A cost-effective image-based model without constraint for weight estimation of poultries using deep learning
The ability to accurately measure body weight phenotypes is paramount to the principles of precision livestock farming. Manual measurement is a time-consuming and labour-intensive process in poultry farming. Consequently, some poultry farms have opted to set up automated measuring stations to measur...
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/S2772375525003843 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The ability to accurately measure body weight phenotypes is paramount to the principles of precision livestock farming. Manual measurement is a time-consuming and labour-intensive process in poultry farming. Consequently, some poultry farms have opted to set up automated measuring stations to measure large numbers of animals. However, these stations are expensive and the weight sensors are less stable. To address these challenges, this study proposes a novel methodology for weight prediction, utilising three angles of camera view in conjunction with a dedicated data processing frame. The proposed methodology integrates a novel lightweight classification model, DeformAttn-ShuffleNetV2, to automatically filter optimal standing posture images, thereby significantly improving the quality of the input data. Furthermore, the Segment Anything Model (SAM) is employed to ensure the accurate segmentation of multiple views, thus facilitating the effective extraction of precise morphological information from complex visual backgrounds that are characteristic of poultry housing conditions. The segmented tri-view images are then entered into the TriWeightNet (Tri-view Weight Estimation Network) model, a lightweight regression model designed to predict continuous body weight values with a high degree of accuracy. The experimental validation process was undertaken on a dataset comprising 13,320 images collected from 275 Pekin ducks under natural farming conditions. The results demonstrated that the mean absolute error of the method was 0.041 kg, representing an 18 % improvement in accuracy compared to conventional weighing approaches. This research provides an innovative solution for practical poultry farming and advances the integration of deep learning and precision livestock management technologies. |
|---|---|
| ISSN: | 2772-3755 |