In vitro fertilisation procedure assisted with computer vision models for organic Senegalese sole (Solea senegalensis) culture

Reproductive dysfunction remains a major challenge for Senegalese sole (Solea senegalensis) aquaculture. Hormone-induced ovulation and in vitro fertilisation (IVF) are currently used to overcome the absence of natural courtship behaviours in captivity. This study investigates the feasibility of hormone-free IVF and behaviour-based prediction of ovulation as alternative strategies to enhance reproductive outcomes. We selected males using computer-assisted sperm analysis to assess sperm motility and quality for IVF trials. IVF trials were conducted using selected males and naturally ovulated eggs collected from females during evening hours across six experimental nights in two groups. Fish behaviour was continuously recorded using underwater cameras, and a convolutional neural network was developed to automatically detect Rest the Head (RTH) and Locomotor Activity (LA) behaviours. These behavioural counts, together with timing information, were used as features to train a logistic regression model for predicting ovulation events. Hormone-free IVF achieved fertilization rates up to 44% with 18% hatching success, producing viable larvae without hormonal intervention. Both groups showed significantly elevated RTH and LA during ovulation nights compared to non-ovulation nights, with peak activity occurring between 18:00-19:00 hours. The behavioural prediction model correctly identified ovulation with 82-85% accuracy and an area under the curve of 0.95. These findings demonstrate that sperm-quality-based male selection combined with automated behaviour analysis provides a practical, non-invasive approach for hormone-free reproduction in organic flatfish aquaculture.