Egg production monitoring in commercial laying cages via the StrongSort-EGG tracking-by-detection model

Abstract

Egg production monitoring provides crucial data on the productivity and health status of laying hens as well as the performance of management practices. In conventional cage systems installed for large-scale farming, the daily egg production of hens in each cage is difficult to record by manual counting. Moreover, the development of automatic counting by robots with an RFID or QR codes on each cage is limited by certain shortcomings, such as the high economic costs and difficulty in installation and maintenance. Therefore, we designed and implemented a novel method for monitoring the number of eggs within the same laying hens cage, and it operates by counting eggs at the cage level. The proposed method is based on the StrongSort-EGG tracking-by-detection model, and it automatically localizes cages by tracking cage columns, counts the egg number, and assigns the number of eggs to a corresponding cage. In addition to the StrongSort-EGG model, the YOLOv5s model was used as the baseline model for detecting eggs and cage columns, and a series of optimization strategies were applied to improve its accuracy and efficiency. The results showed that the optimized YOLOv5s achieved a mean average precision of 99.4 % for cage column targets and 99.3 % for egg targets, with a model size of 5.1 MB and an inference speed of 128.7 frames per second. In a field test involving 200 cages in a commercial laying hen house, the cage localization accuracy was 100 % and cage-based counting accuracy was 98.5 %. Furthermore, the counting time was reduced by more than two-fold compared to that obtained by manual counting. This method provides a new solution for egg counting at the cage level in cage systems and thus may improve management efficiency and reduce potential economic losses.