Implementation of a prototype desktop software based on computer vision for the growth traceability of rainbow trout fish (Oncorhynchus mykiss) in the LESTOMA-UDEC Laboratory

Abstract

Monitoring the growth of rainbow trout (Oncorhynchus mykiss) in closed recirculating aquaculture systems is an essential component for optimizing feed utilization by the fish and enhancing aquaculture production efficiency. This monitoring process requires the application of precise, non-invasive methods that facilitate the assessment of fish development without causing stress and while maintaining the conditions of the aquatic environment. In this study, a computer vision system was implemented to process simultaneous images captured by an overhead camera and a side-view camera positioned above a passage tunnel connecting two aquaculture production ponds. The system estimates the dimensions of sampled fish on a weekly basis using machine learning algorithms and generates data that automatically adjusts the fish feed dosage. To achieve this, the requirements and approximate software design were initially determined while simultaneously building the physical photographic sampling system; then, adjustments were made to image capture thru experimental tests that determined the optimal distance between the cameras and the observation tunnel, evaluating distances from 3 cm to 7 cm. Next, the optical distortion caused by the lens system and the aquatic medium was addressed, for which a polynomial correction function was derived to compensate for the errors present in the captured images. Next, a machine learning algorithm was implemented in Python, and finally, the experimental results were analyzed and validated, demonstrating that at a 7 cm distance between the sampling tube and the camera, the captured images retained the entire fish body within the field of view, preventing cropping at the edges and reducing length estimation errors. Likewise, the execution of the intelligent algorithm and the incorporation of optical correction improved the correlation between the actual measurements and those calculated by the system, as well as generating a response command for the electronic system that periodically controls and adjusts the amount of feed for the fish.