Caged broiler aggregation behavior recognition via target detection and label merging

Abstract

In modern broiler farming, precise environmental control is crucial for the health and production efficiency of the flock, particularly during the chick stage where temperature fluctuations can easily induce stress responses. Currently, intensive farming relies on large-scale environmental control systems for climate regulation and is progressively advancing towards intelligent and precision-oriented development. However, the rapid and accurate assessment of broilers’ adaptability to their environment remains a pivotal challenge. This study proposes an automated detection method based on computer vision to recognize the aggregation behavior of caged broilers. The method employs the YOLOv8-CBAM model to detect individual and group areas, combined with an optimization algorithm based on Relative Intersection Ratio (RIR) to enhance the recognition accuracy of broiler aggregation behavior. Subsequently, by extracting the spatial distribution characteristics of the flock, a random forest classifier is utilized to classify the distribution state into three categories: “Dispersed,” “Normal,” and “Aggregated”. The experimental results demonstrate that the method achieves 94.44% accuracy, 93.88% precision, and 96.67% recall in the recognition of aggregation behavior. Furthermore, in the long video analysis test, the detected trends of flock aggregation and dispersion show precise correspondence with actual observations. This study provides an efficient and intelligent solution for monitoring aggregation behavior of broilers in caged environments, contributing to the realization of precise environmental control and the enhancement of farming management standards.