Multimodal Feature-Driven Deep Learning for the Prediction of Duck Body Dimensions and Weight

TL;DR

This paper tackles non-invasive estimation of duck body dimensions and weight to support precision poultry farming. It proposes a multimodal deep learning framework that fuses 2D multi-view RGB and depth images with 3D point-cloud features, using a modified PointNet++ to extract keypoints and a Transformer encoder to integrate features for regression. The approach achieves a mean absolute percentage error of 6.33% and an R2 of 0.953 across eight morphometric targets on a dataset of 1,023 Linwu ducks, demonstrating strong accuracy while avoiding animal stress. The study introduces the first DL-based poultry morphometrics method, with potential applicability to other poultry species through dataset expansion and improved fusion strategies.

Abstract

Accurate body dimension and weight measurements are critical for optimizing poultry management, health assessment, and economic efficiency. This study introduces an innovative deep learning-based model leveraging multimodal data-2D RGB images from different views, depth images, and 3D point clouds-for the non-invasive estimation of duck body dimensions and weight. A dataset of 1,023 Linwu ducks, comprising over 5,000 samples with diverse postures and conditions, was collected to support model training. The proposed method innovatively employs PointNet++ to extract key feature points from point clouds, extracts and computes corresponding 3D geometric features, and fuses them with multi-view convolutional 2D features. A Transformer encoder is then utilized to capture long-range dependencies and refine feature interactions, thereby enhancing prediction robustness. The model achieved a mean absolute percentage error (MAPE) of 6.33% and an R2 of 0.953 across eight morphometric parameters, demonstrating strong predictive capability. Unlike conventional manual measurements, the proposed model enables high-precision estimation while eliminating the necessity for physical handling, thereby reducing animal stress and broadening its application scope. This study marks the first application of deep learning techniques to poultry body dimension and weight estimation, providing a valuable reference for the intelligent and precise management of the livestock industry with far-reaching practical significance.

Figures (8)

• Figure 1: Collection devices and their layout.
• Figure 2: Capture software.
• Figure 3: Seven annotated feature points on the duck's point cloud used for geometric feature extraction, including key anatomical landmarks such as the beak tip, head peak, neck curve, and tail tip
• Figure 4: Segmentation results of the duck's side-view.
• Figure 5: Architecture of the proposed multimodal model for predicting duck body dimensions and weight. (a) 2D feature extraction: Three independent ResNet50 models extract visual features from top-view RGB, side-view RGB, and depth images, respectively. (b) 3D feature extraction: A modified PointNet++ model performs hierarchical feature extraction from the duck point cloud through three Set Abstraction layers. The final global features are then processed by fully connected layers to predict seven anatomical keypoints in 3D coordinates. (c) Multimodal feature fusion: Image features and 3D geometric features are integrated and refined through a Transformer encoder, enabling accurate prediction by capturing global feature dependencies across different data modalities.