基于A-Unet的牛体尺测量方法研究

Establishment of an A-Unet Based Cattle Body Size Measurement Method

[目的]建立一种基于U-Net改进的A-Unet图像分割与牛体尺测量方法,以实现对牛体体高、体长、体斜长的自动化测量。[方法]首先,在牧场通过摄像头采集牛的侧视图片;其次,利用A-Unet算法进行图像分割,提取牛体边缘轮廓曲线,在牛体轮廓曲线的基础上采用动态网格法寻找牛体尺测量点;最后,根据摄像头已标定的参数和提取到的测量点进行计算,得出牛体尺数据。[结果]通过对深度学习算法图像分割性能的对比分析发现,相比原始的U-Net算法,建立的A-Unet算法具有更高的准确度。利用该算法对牧场21头牛进行体尺指标测定,并与人工测量结果进行比较,经验证,该方法检测体高、体长、体斜长的平均相对误差分别为4.16%、4.05%、4.27%。[结论]基于A-Unet的牛体尺测量方法可以有效地替代传统的牛体尺人工测量方式,具有适用性好、稳定性强和检测准确率高等优点,测量误差能够满足牧户对牛体尺测量需求。

[Objective]This study aimed to establish an improved A-Unet image segmentation and cattle body size measurement method on the basis of U-Net to achieve the automated measurement of cattle body height,body length,and body oblique length.[Method]Firstly,side view images of the farm-raised cattle were collected through cameras.Secondly,the A-Unet algorithm was used for image segmentation to extract the contour curve of cattle body edge.Based on the contour curve of cattle body,the dynamic grid method was adopted to find the cattle body size measurement points.Finally,according to the calibrated parameters by cameras and the extracted measurement points,the cattle body size data was calculated.[Result]The established A-Unet algorithm was found to have higher accuracy than the original U-Net algorithm through comparative analysis of the image segmentation performance of deep learning algorithms.Compared with the manual measurement,the average relative errors of the body height,body length and body oblique length of 21 farm-raised cattle measured by the established A-Unet algorithm were 4.16%,4.05%and 4.27%,respectively.[Conclusion]With the advantages of good applicability,high stability and high detection accuracy,the A-Unet based cattle body size measurement method could effectively replace the traditional manual measurement method.The measurement error met the needs of herdsmen for measuring cattle body size.