| Title |
Application of Deep Learning-based Image Segmentation Algorithm for Korean Cattle Weight Estimation |
| Authors |
이창복(Chang-bok Lee) ; 조현종(Hyun-chong Cho) |
| DOI |
https://doi.org/10.5370/KIEE.2021.70.9.1336 |
| Keywords |
Image segmentation; Korean cattle; Mask R-CNN; Weight estimation |
| Abstract |
Image-based weight measurement is a novel approach for reducing the manpower consumption of Korean cattle farms without causing stress to animals. In this paper, the weight was predicted by measuring the area of Korean cattle and estimating the weight correlation factors. Using the histogram sum from the acquired image data, repeated data was removed and learning efficiency was improved, and the Mask R-CNN model was trained with the selected data. Mask R-CNN uses RoIAlign and is a model capable of precise segmentation by replacing the Softmax activation function of FCN with sigmoid. The result of territorial division of Korean cattle in the background of an actual barn reached an average precision of 0.88 with confidence standards of average IoU 0.88 and 0.85. The size of the mask obtained as a result of segmentation had a maximum correlation of 0.88 with the body weight, and the minor axis length and major axis length of the optimal ellipse calculated through the mask had a correlation of 0.35 and 0.56, respectively. The weighted average of the error rate of the weight prediction results using these three weight correlation factors was 0.07. |