Jian Yue, Pu Shihua, Zhu Jiaming, Zhang Jianlong, Xing Wenwen
Chongqing Academy of Animal Sciences, Chongqing 402460, China.
College of Biosystems Engineering and Food Science, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, China.
Animals (Basel). 2024 Dec 5;14(23):3520. doi: 10.3390/ani14233520.
Controlling the backfat thickness of sows within an appropriate range during different production stages helps to increase the number of pigs weaned per sow per year and ultimately enhances the economic benefit to the pig farm. To obtain the backfat thickness of sows automatically, a backfat thickness estimation method based on machine vision is proposed. First, the backfat thickness values and 3D images of the buttocks of 154 Landrace-Yorkshire crossbred sows were obtained using a veterinary ultrasound backfat meter and Azure Kinect DK camera. After preprocessing the 3D images utilizing Python 3.9.16 software, 10 external morphological parameters reflecting the area, width, height, and contour radius of the sow's buttocks were extracted. The relationships between backfat thickness and external morphological parameters were analyzed in a randomly selected group of 100 sows. A significant positive correlation was observed between backfat thickness and buttock morphological parameters, with the Pearson coefficient for the fitted ellipse area achieving values up to 0.90. A backfat thickness estimation model was developed based on selected buttock feature parameters. The model's generalization performance was evaluated using 54 additional sows that were not involved in the model development. The coefficient of determination (R) between the estimated and actual backfat thicknesses was 0.8923, with a mean absolute error (MAE) of 1.23 mm and a mean absolute percentage error (MAPE) of 5.73%. These metrics indicate that the model can meet production requirements, and the proposed technique offers improved estimation accuracy compared to existing methods. Ultimately, a backfat thickness automatic estimation system was developed using LabVIEW 2023 Q1 (64-bit) software. This research helps to address the cumbersome process of measuring sow backfat thickness and promotes the automation of sow farms.
在不同生产阶段将母猪的背膘厚度控制在适当范围内，有助于提高每头母猪每年的断奶仔猪数量，并最终提高猪场的经济效益。为了自动获取母猪的背膘厚度，提出了一种基于机器视觉的背膘厚度估计方法。首先，使用兽医超声背膘仪和Azure Kinect DK相机获取了154头长白-大白杂交母猪臀部的背膘厚度值和三维图像。利用Python 3.9.16软件对三维图像进行预处理后，提取了反映母猪臀部面积、宽度、高度和轮廓半径的10个外部形态参数。在随机选取的100头母猪中分析了背膘厚度与外部形态参数之间的关系。观察到背膘厚度与臀部形态参数之间存在显著正相关，拟合椭圆面积的皮尔逊系数高达0.90。基于选定的臀部特征参数建立了背膘厚度估计模型。使用另外54头未参与模型开发的母猪对模型的泛化性能进行了评估。估计背膘厚度与实际背膘厚度之间的决定系数（R）为0.8923，平均绝对误差（MAE）为1.23毫米，平均绝对百分比误差（MAPE）为5.73%。这些指标表明该模型能够满足生产要求，与现有方法相比，所提技术的估计精度有所提高。最终，使用LabVIEW 2023 Q1（64位）软件开发了背膘厚度自动估计系统。本研究有助于解决母猪背膘厚度测量过程繁琐的问题，并推动母猪养殖场的自动化进程。
