Exploratory Study of Sex Identification for Chicken Embryos Based on Blood Vessel Images and Deep Learning

• Published in: Agriculture (Basel) Vol. 13; no. 8; p. 1480
• Main Authors: Jia, Nan, Li, Bin, Zhao, Yuliang, Fan, Shijie, Zhu, Jun, Wang, Haifeng, Zhao, Wenwen
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.08.2023
• Subjects: Accuracy; Algorithms; blood vessel; Blood vessels; chicken embryo; Chickens; Data mining; Deep learning; Eggs; Embryonic development; Embryos; Energy consumption; Identification; Identification methods; Image acquisition; Image filters; Image processing; Incubation; Light; Lighting; Machine learning; Neural networks; Poultry; Poultry industry; Sex; sex identification; Sexing; Spectrum analysis; Target detection; Vision systems; VOCs; Volatile organic compounds; YOLOv7; China; Beijing China
• ISSN: 2077-0472; 2077-0472
• DOI: 10.3390/agriculture13081480

The identification of a chicken’s sex is a massive task in the poultry industry. To solve the problems of traditional artificial observation in determining sex, such as time-consuming and laborious, a sex identification method of chicken embryos based on blood vessel images and deep learning was preliminarily investigated. In this study, we designed an image acquisition platform to capture clear blood vessel images with a black background. 19,748 images of 3024 Jingfen No. 6 breeding eggs were collected from days 3 to 5 of incubation in Beijing Huadu Yukou Poultry Industry. Sixteen thousand seven hundred sixty-one images were filtered via color sexing in 1-day-old chicks and constructed the dataset of this study. A sex identification model was proposed based on an improved YOLOv7 deep learning algorithm. An attention mechanism CBAM was introduced for YOLOv7 to improve the accuracy of sex identification of chicken eggs; the BiFPN feature fusion was used in the neck network of YOLOv7 to fuse the low-level and high-level features efficiently; and α-CIOU was used as the bounding box loss function to accelerate regression prediction and improve the positioning accuracy of the bounding box of the model. Results showed that the mean average precision (mAP) of 88.79% was achieved by modeling with the blood vessel data on day 4 of incubation of chicken eggs, with the male and female reaching 87.91% and 89.67%. Compared with the original YOLOv7 network, the mAP of the improved model was increased by 3.46%. The comparison of target detection model results showed that the mAP of our method was 32.49%, 17.17%, and 5.96% higher than that of SSD, Faster R-CNN, and YOLOv5, respectively. The average image processing time was 0.023 s. Our study indicates that using blood vessel images and deep learning has great potential applications in the sex identification of chicken embryos.