Incubation is a key life history stage for birds, and incubation attentiveness can have significant fitness consequences for both parents and offspring. Incubation is, however, a challenging phenomenon to observe and studies generally either measure some proxy of the target behavior, or risk disturbing birds through direct observation. More recently, nest cameras have provided a non-intrusive way to directly observe incubation, but analysis of these data is time-consuming. Here, we use the results of the first deep learning model which automated analysis of nest camera video recordings from eight purple martin (Progne subis) nests over the entire incubation period at a 1-s resolution. We mathematically define the initiation of incubation, characterize the change in nest attentiveness during incubation, and analyze the factors determining nest attentiveness and on- and off-bout duration during the incubation process. A random forest regression model identified the most important predictors of nest attentiveness. Attentiveness decreased with increasing temperature, but the strength of this response increased above the presumed physiological zero egg temperature, below which egg development ceases. This implies that the purple martins are able to adjust their incubation behavior in a complex, multiple-state manner to an extrinsic stimulus. Our study highlights the value of high-resolution datasets created using artificial intelligence for the analysis of nest camera video recordings of animal behavior. The use of artificial intelligence for image classification tasks is becoming commonplace in society. This technology is beginning to be used to automate the analysis of video recordings of wildlife behavior. Here, we use the results of the first such classification from nest camera video recordings of the purple martin (Progne subis) to determine the factors affecting incubation attentiveness (the proportion of time that the adults spend in contact with eggs). Incubation attentiveness is important because it can affect hatch rate and have carry-over effects both for the condition of the incubating adults and the quality of the resulting offspring. Our analysis found that attentiveness was mainly affected by ambient temperature, with incubating adults reducing their efforts as ambient temperature reaches the minimum threshold for egg development.

中文翻译：

---

巢穴摄像机视频记录的深度学习分析揭示了紫色马丁 (Progne subis) 的温度敏感孵化行为

孵化是鸟类生命史的一个关键阶段，孵化的注意力对父母和后代都有显着的健康影响。然而，孵化是一个具有挑战性的观察现象，研究通常要么测量目标行为的某些代表，要么通过直接观察冒着打扰鸟类的风险。最近，巢穴相机提供了一种直接观察孵化的非侵入性方式，但对这些数据的分析非常耗时。在这里，我们使用第一个深度学习模型的结果，该模型在整个孵化期内以 1 秒的分辨率自动分析来自八个紫色马丁 (Progne subis) 巢穴的巢穴摄像机视频记录。我们在数学上定义孵化的开始，表征孵化过程中巢穴注意力的变化，并分析孵化过程中决定巢穴注意力和断断续续时间的因素。随机森林回归模型确定了筑巢注意力最重要的预测因素。注意力随着温度的升高而降低，但这种反应的强度在高于假定的生理零鸡蛋温度时增加，低于该温度鸡蛋发育停止。这意味着紫色马丁能够以复杂的、多状态的方式调整其孵化行为以适应外在刺激。我们的研究强调了使用人工智能创建的高分辨率数据集用于分析动物行为的巢穴摄像机视频记录的价值。将人工智能用于图像分类任务在社会中变得司空见惯。这项技术开始用于自动分析野生动物行为的视频记录。在这里，我们使用紫色马丁 (Progne subis) 的巢穴摄像机视频记录的第一次此类分类的结果来确定影响孵化注意力的因素（成虫与鸡蛋接触的时间比例）。孵化注意力很重要，因为它会影响孵化率，并对孵化成虫的状况和后代的质量产生影响。我们的分析发现注意力主要受环境温度的影响，当环境温度达到卵子发育的最低阈值时，孵化成虫会减少他们的努力。我们使用紫色马丁 (Progne subis) 的巢穴摄像机视频记录的第一个此类分类结果来确定影响孵化注意力的因素（成虫与鸡蛋接触的时间比例）。孵化注意力很重要，因为它会影响孵化率，并对孵化成虫的状况和后代的质量产生影响。我们的分析发现注意力主要受环境温度的影响，当环境温度达到卵子发育的最低阈值时，孵化成虫会减少他们的努力。我们使用紫色马丁 (Progne subis) 的巢穴摄像机视频记录的第一个此类分类结果来确定影响孵化注意力的因素（成虫与鸡蛋接触的时间比例）。孵化注意力很重要，因为它会影响孵化率，并对孵化成虫的状况和后代的质量产生影响。我们的分析发现注意力主要受环境温度的影响，当环境温度达到卵子发育的最低阈值时，孵化成虫会减少他们的努力。孵化注意力很重要，因为它会影响孵化率，并对孵化成虫的状况和后代的质量产生影响。我们的分析发现注意力主要受环境温度的影响，当环境温度达到卵子发育的最低阈值时，孵化成虫会减少他们的努力。孵化注意力很重要，因为它会影响孵化率，并对孵化成虫的状况和后代的质量产生影响。我们的分析发现注意力主要受环境温度的影响，当环境温度达到卵子发育的最低阈值时，孵化成虫会减少他们的努力。