Comparison of CNN-Based Design for Shrimp Seed Counting Machine

Vannamei shrimp is the main commodity for shrimp exports in Indonesia. One of the important processes in shrimp cultivation is the process of sowing the seeds. Among the crucial stages of shrimp farming, the initial step involves purchasing shrimp seeds from nurseries before commencing the seeding process. Currently, the most process of calculating shrimp seeds is still done manually, leading to reduced efficiency and inadequate accuracy. In this research, the authors propose an implementation of a shrimp seed counting machine to enhance the manual counting method. By utilizing a camera, the machine captured images of shrimp seeds to facilitate calculation. Therefore, the obtained image from the camera was processed in the Raspberry PI 4 processor using various CNN methods, i.e., Faster R-CNN ResNet50, Faster R-CNN ResNet101, and SSD MobileNetV2. The primary objective of this research to compare these three models. Based on the F1 score, it is found that ResNet101 achieved the best performance in detection compared to the other two methods. However, this method results in the longest executing time compared to Faster R-CNN ResNet50 and SSD MobileNetV2.