Exploring Pixel Segmentation with Mask R-CNN: Implications for Predicting Cattle Weight

• Published in: 2023 3rd International Conference on Smart Cities, Automation & Intelligent Computing Systems (ICON-SONICS) pp. 32 - 37
• Main Authors: Yulianingsih, Nurdiati, Sri, Sukoco, Heru, Sumantri, Cece
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06.12.2023
• Subjects: Agriculture; Cows; Feature extraction; Image segmentation; livestock; Mask R-CNN; Object detection; pixel; prediction; Predictive models; Reliability; segmentation
• DOI: 10.1109/ICON-SONICS59898.2023.10435120

In the realm of livestock management, particularly concerning cattle, the swift and precise estimation of animal weight is of paramount significance. This research delves into the intricacies of image data preprocessing utilizing various combinations of the Mask R-CNN approach to support deep learning-driven cattle weight prediction models. The renowned efficacy of Mask R-CNN in object detection and segmentation is harnessed to emphasize features specific to cattle. Rigorous experiments were undertaken, encompassing 31 distinct datasets, aggregating a total of 223 images of Bali cattle before augmentation. The empirical findings emphasize the superiority of the Mask R-CNN R50-DC5 and R50-FPN backbone combinations, which exhibited commendable precision rates of 86,1% for R50-DC5. 85,8% for R50-FPN, and 85,1% for R50-C4. Additionally, observations were conducted to ascertain the most efficient process speed among these configurations. The manuscript aims to select the optimal image segmentation method to acquire pixels that will be explored as features in predicting the weight of livestock. The results obtained underscore the importance of ensuring meticulous evaluations of image segmentation accuracy to maintain the integrity and reliability of the model.