Improving Shrimp Seed Counting Performance Using YOLOv5 and Super Resolution Method

This paper presents the development of a shrimp seed detection method on a previously existing shrimp seed counting device. The proposed method can count a larger number of shrimp seeds in a shorter amount of time compared to the previous method. We used the YOLOv5 method to detect shrimp seeds. However, since YOLO has weaknesses in detecting very small objects, we improved YOLO's ability by dividing the detection area into several smaller parts, and then enlarging each part to a certain size by involving superresolution techniques. With the method we used, the test results showed an accuracy of 98.14% on a number of 150, 3-day-old shrimp seeds. In addition, we also succeeded in speeding up the detection computation process to 90.75 seconds.