Abstract:

High hatchery yields are achieved in chicken egg incubation， however， research on the detection of dead embryos and infertile eggs at different incubation time is scarce. During the early incubation period， due to the weak embryo activity and low heat production， thermal imaging detection struggles to effectively differentiate live embryos from dead embryos and infertile eggs； from the middle to later stages of incubation， as live embryos become more developed， light cannot penetrate through the eggs， rendering optical detection methods ineffective in distinguishing dead embryos. To address these issues， optical detection and thermal imaging detection was combinated to identify dead embryos and infertile eggs during the early （6 to 12 d ） and mid-incubation periods （15 to 18 d）. Changes in light transmission of eggs during 6 to 12 d were captured using a photoelectric sensor， and feature parameters were extracted. Additionally， thermal images of eggs’ natural cooling temperature drop sequences during 15 to 18 d were collected using custom-developed software. After obtaining region of interest templates after image processing， the sequence of thermal images was segmented， and the grayscale mean change curves of egg regions were derived. Feature parameters reflecting egg viability were extracted， and detection models for different incubation time points were established using logistic regression （LR）， K-nearest neighbors （KNN）， support vector machine （SVM）， decision tree （DT）， naive Bayes （NB） and random forest （RF） algorithms. Results indicated that KNN performed best in distinguishing egg viability from 6 to 12 d， with an accuracy of 98.7% on 10 d. LR demonstrated the most stable performance across all days， achieving an accuracy of 94% in distinguishing egg viability from 6 to 12 d. SVM performed best in distinguishing egg viability from 15 to 18 d， with an accuracy of 97.1% at 18 d. Utilizing optical detection to detect changes in light transmission during the early incubation period and thermal imaging technology to analyze grayscale mean temperature changes of eggs during the mid-incubation period enabled batch detection of egg viability in different stages of incubation for White Leghorn chicken eggs.

Key words: eggs, photoelectric detection, thermal imaging, image processing, fertility of hatching eggs

摘要：

鸡蛋孵化产量高，但不同孵化时间节点死胚蛋与无精蛋的检测研究较少。孵化前期由于胚蛋活性较弱、产热较少，热成像检测不能很好地将活胚蛋从死胚蛋和无精蛋中区分出来；孵化中后期由于活胚蛋发育基本成型，光线不能贯穿种蛋，光电检测方法无法区分中后期死亡的胚蛋。将光电检测与热成像检测结合，识别孵化前期（6~12 d）与孵化中期（15~18 d）死胚蛋与无精蛋。通过光电传感器采集孵化6~12 d种蛋的透光量变化，提取特征参数，使用自主开发的软件采集孵化15~18 d自然冷却温降序列的热图像，经过处理获取感兴趣区域模板，利用模板分割序列热图像，获得种蛋区域灰度均值变化曲线，提取反映种蛋活性的特征参数，使用逻辑回归（logistic regression，LR）、K最近邻（K-nearest neighbors，KNN）、支持向量机（support vector machine，SVM）、决策树（decision tree， DT）、朴素贝叶斯（naive Bayes， NB）、随机森林（random forest，RF）建立不同孵化时间节点的检测模型。结果表明，KNN区分孵化第6~12天种蛋活性效果最好，在孵化第10天准确率达98.7%；LR区分第6~12天种蛋活性效果在所有时间最稳定，准确率均达94%。SVM区分孵化第15~18天种蛋活性效果最好，第18天准确率达97.1%。孵化前期使用光电检测利用透光量变化检测死胚蛋与无精蛋，孵化中期使用热成像技术利用种蛋温度变化的灰度均值，提取灰度变化曲线特征检测死胚蛋，可批量检测不同孵化阶段白壳白羽鸡种蛋活性。

关键词: 鸡蛋, 光电检测, 热成像, 图像处理, 种蛋活性