10:15 〜 10:30
[XYS-19-06] Study on estimating method of feed intake of broilers based on object detection by deep learning
Objective
In this study, we aimed to establish a management method to grasp feed intake of broilers based on image recognition by artificial intelligence (AI). We examined whether using object detection by deep learning could estimate feed intake of broilers.
Methods
The experimental animal were broilers (chunky, female). For making AI, object detection by deep learning was used. Model A, B, C, and D were made from the training datasets prepared by 800, 400, 200, 100 images of broilers with eating behavior. For accuracy evaluation of AI, a test dataset was used which consisting of 100 images of broilers with eating behavior and 100 images of broilers without any eating behavior. The indicator of accuracy evaluation was F1 score. For feed intake estimation, the correlation (Simple regression analysis) between the number of eating behavior detected in the eating behavior’s videos by the AI with the highest detection accuracy and the corresponding feed intake was investigated.
Results
The AI for eating behavior detection (Model A, B, C, D) 's F1 score was 0.99, 0.97, 0.94, 0.85, respectively. The coefficient of determination R2 by the simple regression analysis was 0.85.
In this study, we aimed to establish a management method to grasp feed intake of broilers based on image recognition by artificial intelligence (AI). We examined whether using object detection by deep learning could estimate feed intake of broilers.
Methods
The experimental animal were broilers (chunky, female). For making AI, object detection by deep learning was used. Model A, B, C, and D were made from the training datasets prepared by 800, 400, 200, 100 images of broilers with eating behavior. For accuracy evaluation of AI, a test dataset was used which consisting of 100 images of broilers with eating behavior and 100 images of broilers without any eating behavior. The indicator of accuracy evaluation was F1 score. For feed intake estimation, the correlation (Simple regression analysis) between the number of eating behavior detected in the eating behavior’s videos by the AI with the highest detection accuracy and the corresponding feed intake was investigated.
Results
The AI for eating behavior detection (Model A, B, C, D) 's F1 score was 0.99, 0.97, 0.94, 0.85, respectively. The coefficient of determination R2 by the simple regression analysis was 0.85.