Abstract:

[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.

Key words: image segmentation, cattle body size measurement, A-Unet, image processing