Image-Based Shrimp Aquaculture Monitoring

Shrimp farming is a growing industry, and automating certain processes within aquaculture tanks is becoming increasingly important to improve efficiency. This paper proposes an image-based system designed to address four key tasks in an aquaculture tank with Penaeus vannamei: estimating shrimp length and weight, counting shrimps, and evaluating feed pellet food attractiveness. A setup was designed, including a camera connected to a Raspberry Pi computer, to capture high-quality images around a feeding plate during feeding moments. A dataset composed of 1140 images was captured over multiple days and different times of the day, under varying lightning conditions. This dataset has been used to train a segmentation model, which was employed to detect and filter shrimps in optimal positions for dimensions estimation. Promising results were achieved. For length estimation, the proposed method achieved a mean absolute percentage error (MAPE) of 1.56%, and width estimation resulted in a MAPE of 0.15%. These dimensions were then used to estimate the shrimp's weight. Shrimp counting also yielded results with an average MAPE of 7.17%, ensuring a satisfactory estimation of the population in the field of view of the image sensor. The paper also proposes two approaches to evaluate pellet attractiveness, relying on a qualitative analysis due to the challenges of defining suitable quantitative metrics. The results were influenced by environmental conditions, highlighting the need for further investigation. The image capture and analysis prototype proposed in this paper provides a foundation for an adaptable system that can be scaled across multiple tanks, enabling efficient, automated monitoring. Additionally, it could also be adapted to monitor other species raised in similar aquaculture environments.