Reducing costs in sturgeon aquaculture: AI-assisted sex determination and strategic innovations

Justification

Sturgeon aquaculture is a significant sector within the Western regional aquaculture industry, primarily driven by the high value of sturgeon products such as caviar and protein. However, the process of sex determination, crucial for maximizing economic returns from each fish, currently represents an operational bottleneck. Traditional methods, such as ultrasound and surgical biopsy, are not only labor and time-intensive but also subject the fish to stress due to the need to capture, dewater, and sedate the fish. Given the potential 100,000 to a multiple millions of sturgeon managed annually on farms, these methods significantly escalate operational costs. This is compounded by labor requirements and the extended time needed to rear fish to sexual differentiation, further complicated by geographical variability in strain and water temperatures between CA and ID. These challenges underscore the need for a method that can accurately and quickly determine sex months to potentially years sooner, and more efficient than ultrasound, thus optimizing the production cycle and reducing unnecessary expenditures on feed and labor.

Proposed AI Solution: In response to these challenges, we propose an Artificial Intelligence (AI)-assisted approach using advanced machine learning algorithms and computer vision technology to non-invasively determine the sex of young sturgeons. The overall goal of this project is to decrease producer costs associated with sex determination across sturgeon aquaculture in the Western United States using image classification software that leverages machine learning. By analyzing images of the ventral surface of the fish (Fig 1), our solution aims to identify sex much earlier than current methods allow, with minimal human intervention. This work builds on previous work done in machine learning for fish classification using computer vision and automation in aquaculture (Aziz et al., 2023; Costa et al., 2013; Garcia-d’Urso et al., 2022; Gladju et al., 2022; Li et al., 2023; Saleh et al., 2022; Viazzi et al., 2015). We propose using machine learning and automation through computer vision technology to automate the sex determination processes in sturgeon.This early determination will enable producers to streamline their operations by quickly segregating fish based on their sex, thereby reducing the time and resources spent on rearing males primarily used for meat production, which can be brought to market sooner.

Objectives

Objective 1. Develop an AI model capable of rapid sex determination in juvenile sturgeon. Compile and compress efficient model for deployment on edgeAI compute device for real-time rapid sex determination in juvenile sturgeon with >80% test accuracy. Design prototypes for sorting and binning of sturgeon using predictions from the model. 

Objective 2. Establish real time model updates during on farm sex determination activities via a networked feedback loop of individual monitoring of sturgeon, data collection, updating classification, retraining of models and over the air deployment to edgeAI compute devices with >90% test accuracy. Test prototype platform for sorting and binning of sturgeon at commercial sturgeon farms in California.

Objective 3. Increase throughput technology via automation, electronic tracking and AI, as well and provide cost comparison associated with sexing and binning of sturgeon in aquaculture against current methods. Training, extension and outreach objectives from the Outreach and Evaluation Plan objectives 1-3.