基于多源图像和环境信息融合的规模化养殖蛋鸡体温测量方法

Temperature Measurement Method for Commercially Farmed Layer Hens Based on Multi-source Image and Environmental Data Fusion

摘要:

规模化蛋鸡养殖一直以来都面临着蛋鸡健康状态不易评估、疫病无法有效预防等问题，鸡群健康监测对于蛋鸡养殖业的意义日渐显著。蛋鸡作为恒温动物，其体温是评估健康状态的重要指标。本研究以叠层笼养蛋鸡为研究对象，提出了一种融合多源信息的蛋鸡体温测量方法。首先对热红外相机进行温度漂移校正和距离校正，以提高相机的测量精度。将热红外图像与采集的近红外图像和深度图像进行像素级配准，使用YOLO v8n目标检测网络对融合的多源图像进行蛋鸡头部检测，检测结果AP50为97.0%，AP50-95为76.1%。然后根据环境温度和蛋鸡头部距离信息对蛋鸡头部热红外图像进行温度漂移校正和距离校正，提取校正后图像的温度特征点计算蛋鸡头部温度。基于环境温度、环境相对湿度、环境风速、光照强度和蛋鸡头部温度构建了蛋鸡体温预测数据集，利用机器学习算法预测蛋鸡体温。其中随机森林算法在蛋鸡体温预测中表现最好，R2为0.696，RMSE为0.246℃。本研究为实现准确、无扰动地测量规模化蛋鸡养殖场的鸡只体温提供了参考。

Abstract:

Large-scale egg farming faces challenges in assessing the health status of laying hens and preventing disease outbreaks. The need for effective flock health monitoring in egg production is becoming increasingly important. As homeothermic animals, the body temperature of laying hens serves as a crucial indicator of their health. A method for measuring the body temperature of stacked cage laying hens was proposed by integrating multi-source information. To improve measurement accuracy, temperature drift correction and distance correction were applied to the thermal infrared camera. The thermal infrared images were then pixel-level aligned with the acquired near-infrared and depth images. These fused multi-source images were used to detect the heads of the laying hens through the YOLO v8n detection network, achieving detection results of 97.0% for AP50 and 76.1% for AP50-95. Temperature drift and distance corrections were performed on the thermal infrared images of the hens’ heads, using ambient temperature and distance information. Temperature feature points were then extracted from the corrected images to calculate the head temperature of the laying hens. A prediction dataset was constructed based on environmental factors such as ambient temperature, humidity, wind speed, light intensity, and the hens’ head temperature. Various machine learning algorithms were used to predict the body temperature, with the random forest algorithm showing the best performance, achieving an R2 of 0.696 and an RMSE of 0.246℃. The research result can provide a reference for achieving accurate, high-throughput, and non-invasive measurement of body temperature in large-scale egg farms.