Integrating machine learning and infrared smart cameras into critically endangered bird production

Artificial incubation and nest management are useful tools for the conservation of rare, threatened, and endangered bird species. It can reduce disease transmission from parent to chick, stimulate re-nesting by parent birds who double their reproductive output, and allow medical support during hatching to maximize hatchling survival. In some cases, eggs are left in the original or foster nest until just before hatching, then removed, sanitized, and placed in an incubator. The problem is that egg-laying does not occur on a set schedule so the laying date may not be known, and different eggs in the same nest may hatch on different dates. To date, predicting hatch requires frequent nest checks by hand, causing stress to parental birds and potentially decreasing their reproductive output. Here we describe a method to replace manual nest box checks with a machine vision system to automatically alert managers when hatching is imminent. To test this method, thermal imaging photos of eggs (n=42) were taken across 15 species of Psittaciformes (parrots) and Anseriformes (ducks) at a specialized bird breeding facility. A machine learning model produced in Google Vertex AI, detected eggs within 1.5 days of hatch with 100% accuracy, without false negatives. This method has great potential to improve captive breeding or wild nest box monitoring outcomes for rare, threatened, and endangered bird species.