Pig weight estimation using a movable top-mounted depth camera with improved EfficientNetV2
Year: 2025
Type: article
Abstract: Accurate pig weight estimation is crucial for optimizing feed allocation, health management, and slaughter timing. However, most existing methods depend on data collection in constrained environments, limiting their applicability in free-range farming. This study proposes a non-contact estimation system tailored for free-range settings, employing a movable, top-mounted depth camera to capture depth images without restricting pig movement. To enhance estimation accuracy, an improved EfficientNetV2 model is introduced, incorporating the CBAM (Convolutional Block Attention Module) to strengthen spatial feature extraction and adapt to varying pig body shapes. Additionally, multiple pooling strategies in the prediction head are evaluated, and the optimal configuration is selected. The proposed model achieves an MAE of 3.263 kg, with only 5.920 M parameters and 0.752 GFLOPs, outperforming existing approaches. Limitations include potential occlusions in densely populated scenes. Future research will explore multi-view data integration and further improve robustness under complex farm conditions.
Source: Smart Agricultural Technology
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: gold
Estimated APC paid: $1,770
Awards: G5545500390, G6412133804, G7081374565
OpenAlex ID: W4412611237
