AI Powered Sensing to Detect Viable and Early Dead Chick Embryos based on Selective Band Images

The poultry hatchery industry demands an innovative solution to detect dead chick embryos to improve hatchery operations while ensuring the production of healthy chicks. Approximately 5–10% of embryos die due to thermal shock and inappropriate handling during the early incubation stages, posing significant challenges in hatchery practices, including contamination risks, space constraints, manpower and energy costs. To address this issue, a non-destructive approach using hyperspectral imaging system (HSI) combined with deep learning algorithms (YOLO11 and InceptionNet) were used to detect and classify viable and dead chick embryos at day 4. A hyperspectral imaging (HSI) camera (PIKA-L) sensitive to visible-near infrared wavelengths (400–1000 nm) with the scanning speed of 0.06 cm/s with frame rate of 9.9 fps and exposure time of 100 ms were set for egg image acquisition. Half of the incubated eggs were treated with CO ₂ for six hours, during which time they were placed in a cooler at approximately 4 ^o C, at incubation day 2 to induce embryonic death. It has been hypothesized the spectral differences of hemoglobin (Hb) absorbance in the range of 550–580 nm should be varied between live and dead embryos. Therefore, 3 wavelengths near Hb absorbance were selected to get pseudo color images from HSI cube to visualize embryonic circulatory system. On the validation set, the YOLO11 model achieved higher accuracy such as mAP50 = 0.995 and mAP50-95 = 0.995. This research can be used for image based viable chick embryo detection at industrial settings in the future.