การพัฒนาเทคนิคการวัดขนาดปลานิลแบบไม่สัมผัสโดยใช้สเตอริโอวิชั่นและโครงข่ายประสาทเทียม

Abstract

Tilapia cultivation in a closed environment, such as plastic or cement ponds, is extremely effective and efficient. One crucial step is to continuously record the growth data of the tilapia to determine the appropriate amount of feed, antibiotics, and oxygen. However, measuring the size of the tilapia by catching them directly from the breeding pond can greatly affect the fish, causing stress, loss of appetite, or even death. This independent study aims to develop non-contact tilapia measurement technology using the following steps: capturing independently floating tilapia using a second stereo camera system, using an object detection model to separate the tilapia from the background image, using Yolo v8 to locate the position of the tilapia in the image, using stereo vision technology to convert the coordinates of the tilapia in the image into actual coordinates with the camera, and finally calculating the size of the tilapia based on the actual coordinates. After the process is completed, the size of the tilapia is measured in millimeters. The experiments can be divided into Experiment 1, which measures the width and length of the tilapia in a tilapia scene. The R2 values for width and length measurements are 0.993 and 0.998, respectively. Later, when considering the measurement range above, the average MAPE values for width and length are 2.48% and 2.29%, respectively, followed by actual size measurement of the tilapia. The R2 values for actual size measurement are 0.998 and 0.995, respectively. The average MAPE values are 5.82% and 6.96%, respectively. This indicates that this technology can be used to measure the size of future farm fish.