A computer vision and RFID fusion-based method for measuring individual feed intake and its application for detecting individual differences in feed efficiency of large yellow croaker (Larimichthys crocea)

Estimating the individual feed intake (FI) for multiple consecutive meals of fish reared at commercial densities has long been a challenge and this difficulty has prevented the genetic improvement of feed efficiency (FE) in fish. We propose an automatic and real-time measurement system for individual FI of fish reared in a group based on computer vision and radio frequency identification fusion technology in large yellow croaker (Larimichthys crocea). To achieve this, we designed a feeding station where only one fish at a time can enter and have their passive integrated transponder (PIT) tag recorded. We then trained a feed pellet detection model based on You Only Look Once v5 using an annotated dataset, which achieved a final precision of nearly 100%. Finally, we utilized the trained feed detection model combined with PIT scanning to accurately and automatically track individual FI of fish with access to the feeding station. In 10 experiments lasting a total of 792 ​min conducted in the laboratory, the automatic real-time feed counting achieved an average accuracy of 94.5%. In addition, during a 14-day FI measurement period conducted in an indoor farm with 894 fish that received two meals per day, large yellow croaker feed efficiency ratio (FER) was 0.9 ​± ​0.4 with a coefficient of variation of 47%. FER showed a weak positive correlation with initial body weight and a weak negative correlative with FI. There was also a moderate correlation between FER and body weight gain (BWG), with subgroups that had high BWG exhibiting greater FER values. The approach described here demonstrates a method to automatically and accurately investigate FER in fish that can be used to assess the potential for their genetic improvement.