Zhenghao Li , Yinfa Yan , Junjie Zhao , Mingrui Shang , Zhanhua Song , Jing Wang , Tingju Sun , Guizheng Zhang , Mochen Liu 

Abstract

With the adoption of whole-age artificial diet–based rearing in China’s sericulture industry, accurately capturing silkworm feeding behavior has become essential for intelligent farming management. Feeding behavior reflects physiological status and ingestive rhythm, yet current studies remain limited and traditional observations rely on manual experience, preventing continuous and precise characterization. This study proposes a non-invasive quantitative framework that integrates machine vision with multi-scale signal analysis to analyze feeding behavior under artificial-diet conditions. A YOLOv12 and ByteTrack pipeline is used for high-precision detection and continuous tracking of the silkworm head. A dual-coordinate analytical model combining Cartesian and dynamic polar coordinates is constructed to extract high-resolution motion trajectories. Kernel Density Estimation, Fast Fourier Transform, Time-domain analysis, and wavelet transform are applied to reveal multi-scale motion patterns and rhythmic features. Experiments were conducted on 4th instar silkworms at the post-ecdysis initial feeding stage, and the results showed that their feeding behavior presents a stable three-phase feeding rhythm structure composed of the rapid micro-swing phase, slow-rhythm adjustment phase and rest phase: a rapid micro-swing phase representing fine ingestive actions, a slow-rhythm adjustment phase involving directional and postural modulation, and a rest phase reflecting brief physiological pauses. Using the dual-coordinate system analysis model, this study achieved the decoupling of macroscopic movement paths and microscopic movements. The Cartesian coordinate system revealed the dominant gnawing rhythm at 1.5–2.2 Hz and the rapid micro-swings within the feeding zone, while the dynamic polar coordinate system demonstrated the high stability of the radial distance (87% of energy below 0.2 Hz) and the 1–2 Hz periodic bimodal swings of the polar angle around ± 90°. Finally, supplementary extended validation was performed on 3rd and 5th instar silkworms under artificial diet rearing conditions, which confirmed the cross-instar stability of the three-phase structure in 3rd to 5th instar silkworms reared on artificial diet, and clarified the variation trends of core parameters across different instars. This framework enables dynamic, fine-scale quantification of silkworm feeding behavior and provides new evidence for understanding ingestive mechanisms. It also offers a reusable technical pathway for automated behavioral monitoring and intelligent optimization of feeding strategies, and has important application prospects for promoting the intelligent and refined development of industrialized artificial diet rearing of silkworms.

Page Linkage:https://doi.org/10.1016/j.compag.2026.112053%20