Facial Chick Sexing: An Automated Chick Sexing System From Chick Facial Image

TL;DR

This work investigates non-invasive day-old chick sexing by transferring human facial gender recognition techniques to poultry. It presents an end-to-end facial chick sexing system combining YOLOv5-based face localization, HRNet-based keypoint detection, precise face alignment, and CNN-based feature extraction with a three-layer classifier, evaluated with 5-fold cross-validation on a 5,727-image dataset. ResNet-50 consistently yields the best performance across two face crops, achieving around 81–82% accuracy and offering interpretable Grad-CAM++ visualizations that focus on beak and comb areas as discriminative cues. The study demonstrates the practicality of automated, welfare-friendly chick sexing and discusses integration with vaccination hardware and the need for larger, more diverse datasets to further boost robustness and real-time applicability.

Abstract

Chick sexing, the process of determining the gender of day-old chicks, is a critical task in the poultry industry due to the distinct roles that each gender plays in production. While effective traditional methods achieve high accuracy, color, and wing feather sexing is exclusive to specific breeds, and vent sexing is invasive and requires trained experts. To address these challenges, we propose a novel approach inspired by facial gender classification techniques in humans: facial chick sexing. This new method does not require expert knowledge and aims to reduce training time while enhancing animal welfare by minimizing chick manipulation. We develop a comprehensive system for training and inference that includes data collection, facial and keypoint detection, facial alignment, and classification. We evaluate our model on two sets of images: Cropped Full Face and Cropped Middle Face, both of which maintain essential facial features of the chick for further analysis. Our experiment demonstrates the promising viability, with a final accuracy of 81.89%, of this approach for future practices in chick sexing by making them more universally applicable.

Figures (11)

• Figure 1: Data acquisition and annotation process. Our acquisition process involves setting up three cameras for data recording and pre-processing to eliminate low-quality frames and videos. The annotation process is semi-automated, starting with manual annotations to create an initial dataset, followed by model training, prediction and iterative corrections to efficiently annotate the entire dataset.
• Figure 2: Comparison between the two frames on the left that represent an example of frames that do not meet the standards and the one on the right that represents of a frame that satisfies the standards
• Figure 3: Schematic of the gender classification procedure for day-old chicks
• Figure 4: Chick Facial Pose Estimations: Cases of Roll, Pitch, and Some Instances of Yaw
• Figure 5: Overall procedure of cropping to obtain Cropped Middle Face from the Full Face.