Chaos identification for dairy cow weight time series with deep wavelet transform☆

Monitoring the health status of dairy cows is essential for disease prevention and control (Fournel et al., 2017). Information technology has played a significant role in monitoring cattle diseases through the use of modern technologies like computers and sensors (Gertz et al., 2020). This technology-based approach offers advantages such as handling large data volumes and achieving high detection accuracy. It enables precise monitoring of cattle health and disease occurrence, leading to improved disease prevention and control in dairy farming (Hansen et al., 2018).

In previous studies (Ferrero et al., 2023), methods for monitoring the health of cattle mostly involved analyzing images and video footage to detect the condition of the cows. However, the accuracy of recognition was affected by factors such as occlusion and lighting during the feature extraction process, which in turn impacted the prediction of cattle diseases. In recent years, there has been an increasing interest among researchers in analyzing weight signals as a means of monitoring cattle health. Due to the multi-scale nature of weight signals, they can be used to monitor the occurrence and progression of various diseases.

Chaotic analysis methods have proven highly effective for handling complex data across various domains. For instance (Hariri-Ardebili and Mahdavi, 2023), applied chaos theory to analyze the nonlinear behavior of brittle materials like concrete, revealing its value in identifying and quantifying uncertainties and complexities in concrete performance predictions (Zelinka, 2016). highlighted the significance of hidden attractors in chaotic systems, which is crucial for understanding and predicting their dynamic behavior (Ewees and Elaziz, 2020). introduced the Chaotic Multi-Verse Harris Hawks Optimization (CMVHHO) algorithm, enhancing the Multi-Verse Optimizer's (MVO) search and exploration capabilities through chaotic maps, and demonstrated its efficiency in tackling complex engineering optimization challenges (Lin et al., 2011). explored the synchronization of two dissimilar chaotic systems using adaptive type-2 fuzzy sliding mode control, which is vital for applications such as secure communications, biological system modeling, and physical process control.Other studies, such as (Das et al., 2014) investigated a novel swarm dynamics based on signal chaos analysis (Przystałka and Moczulski, 2015). present a methodology for fault detection based on signal chaos analysis (Xia et al., 2022). introduces an identification approach, emphasizing the significance of signal analysis and chaos dynamics in the identification process.

Through chaos analysis of body weight signals, we can explore the chaotic properties of weight signals in cows and further uncover important information related to their health status. Based on these insights, this study proposes a nonlinear dynamical model for bovine leg dynamics, elucidating the principles, mechanisms, and patterns underlying chaotic phenomena. Furthermore, an effective classification technique for detecting anomalies in cattle is introduced, based on deep wavelet transform. The contributions of this study are as follows:

• (1)
  We constructed a nonlinear dynamic model, confirmed the existence of its strong chaotic behavior, and analyzed the chaotic mechanism of the cow weight signal.
• (2)
  By analyzing the Lyapunov exponent, we verified the chaotic nature of the collected weight signals.
• (3)
  We introduced a weight signal anomaly detection model based on deep wavelet transform.
• (4)
  In the detection model, we enhanced the Center-symmetric Local Binary Patterns (CLBP) algorithm as a feature generation function to extract chaotic features.

This paper is organized as follows. Section 2 introduces the work related to dairy cow health monitoring. Section 3 introduces the structure and advantages of the proposed anomaly detection method. The experimental results and analysis are presented in detail in Section 4. Finally, conclusions are drawn in Section 5.