Dam behavior patterns in Japanese black beef cattle prior to calving: Automated detection using LSTM-RNN

https://doi.org/10.1016/j.compag.2019.105178Get rights and content

Highlights

  • LSTM-RNN model was used to classify dam behavior patterns related with calving using IMU sensor.

  • Seven behavior patterns related with calving were classified.

  • Activity level between precede day and calving day were compared by computing VeDBA and ODBA value.

  • The LSTM-RNN models achieved their best performance when using window-sizes 32.

  • ODBA and VeDBA value in calving day were fluctuating more intensive than precede days.

Abstract

This study develops a recurrent neural network (RNN) with a long short-time memory (LSTM) model to detect and recognize calving related behaviors using inertial measurement unit (IMU). The models were trained using IMU data collected from three expectant cows during the last three days before calving. Classified behavior pattern classes included feeding, ruminating (lying), ruminating (standing), lying normal (collected during 72 h–24 h before calving), standing normal (same as lying normal), lying final and standing final, which were defined as the lying and standing behavior that occurred during the last 24 h before calving. The LSTM-RNN models were trained to classify cow behavior classes across window-size of 32, 64, 128 and 256 respectively (1.6 s, 3.2 s, 6.4 s and 12.8 s). The best overall performing LSTM-RNN model had a window-size of 32 (accuracy, precision, recall, f1-score were 79.7%, 81.1%, 79.7% and 79.8%, respectively). With a window-size of 32, the model classification accuracy for specific behaviors was 76.0% (feeding), 92.6% (ruminating (lying)), 88.3% (ruminating (standing)), 63.2% (lying normal), 78.0% (standing normal), 74.7% (lying final) and 70.1% (standing final). These results demonstrate the potential of a LSTM-RNN model to automatically recognize behaviors patterns prior to birth. In the future, more related indicators will be added to improve the accuracy and robustness of this recognition model. With further work, statistically significant changes in behavior could be streamed to farmers informing them of the progress of calving and alerting them to critical changes in the situation.

Introduction

Dystocia, abnormal or difficult birth, is a common occurrence in beef cattle calving, which can have a serious impact on both the calf and dam. For the calf, dystocia increases the risk of stillbirth, calf mortality and morbidity (Lombard et al., 2007). For the dam, it also increases the likelihood of trauma and the risk of uterus and placenta disorders (Schuenemann, 2012). To date, long periods of observation by the farmer just prior to calving have been required in order to ensure timely human assistance and intervention, since calving is predicted on a calendar basis; a period of between 267 and 295 days after successful insemination (Inchaisri et al., 2010). A system that could automatically monitor pregnant dams, and alert farmers and veterinaries to abnormal behavior or imminent calving would significantly reduce the burden of providing timely animal welfare and inform precision farm management.

Several behavioral changes have been observed and documented in the lead up to calving. For example, a decrease in feeding and ruminating from two weeks prior to calving (Bar et al., 2010), while on the day of calving, dams spent less time lying, and there were obvious increases in lying /standing transitions. Standing is one of the main postures during the 12 h prepartum period (Houwing et al., 1990). Moreover, several studies have mentioned that dams become more active in the days preceding calving (Miedema et al., 2011a, Jensen, 2012). Signs of restlessness and irritation also increase sharply in the last 2–6 h preceding calving.

There have been attempts to reduce the burden of monitoring the dam in the lead up to calving using sensors, such as using automatic feeding bins to detect feeding and ruminating behavior, electric halters (Benaissa et al., 2019) and Insentec system (Program and Systems, 2013), or 3-axis accelerometer (Benaissa et al., 2017) or pedometers attached to the cow’s hind leg to detect lying, standing, lying/standing transition and other activity in comparison to behavior during normal pregnancy (especially in last 6 h before calving (Jensen, 2012, Borchers et al., 2017, Saint-dizier and Chastant-maillard, 2015)). To our knowledge, though, these devices are based either on a limited number of behaviors, or manually record and analyze behavior during the last hours of calving. Moreover, most of these studies have been conducted on Holstein dairy cattle; beef cattle have rarely been the focus of these studies. Therefore, in this study, a monitoring system to classify and automatically detect multiple behaviors associated with calving was developed. These behaviors included feeding, ruminating (while lying and standing), lying and standing measured using an inertial measurement unit (IMU) sensor and a Long Short Time Memory-Recurrent Neural Network (LSTM-RNN) model. The LSTM-RNN model is a particular type of recurrent neural network designed to counter the effect of diminishing gradients through layers and suitable for time series data. In this study, LSTM-RNN models were trained with 4 window-sizes in order to evaluate the effect of window-size on classification performance. Moreover, in order to document any change in cow activity levels in the days leading up to calving day, IMU data was automatically recorded over this period. Activity levels from 2 days prior and up to the day of calving were analyzed by computing vectorial dynamic body acceleration (VeDBA) and dynamic body acceleration (ODBA) of standing and lying separately (Miwa et al., 2015). This study aims to offer an effective tool for classification and automated detection of multiple behaviors leading up to actual calving and the prediction of calving time.

Section snippets

Location and animals

The experiment was conducted at the Livestock Farm of Kyoto University, Kyotanba, Kyoto Pref. Japan (35˚19′04.93″N, 135 ˚ 41′67.96″E) between the 10 and 17 January 2018. The IMU data was collected from 3 expectant Japanese Black Beef Cattle (age: 2–9 years). The three dams gave birth to a calf on the 14th, 15th and 16th January 2018, respectively. One of them experienced dystocia caused by a malposition of the calf. The three pregnant cows were placed in two spacious barns in pairs (4 cows were

Results

The confusion matrix results (Fig. 4) show the accuracy of the behavior pattern classifications using the LSTM-RNN model. The accuracy of the seven classified behaviors among the 4 different window-sizes are shown. Generally, in all window-sizes, the behaviors, including feeding and ruminating while lying, ruminating while standing, were classified with a relatively high accuracy and were not affected by window-size. Especially, the accuracy of ruminating while lying and ruminating while

Discussion

In this study, multiple behaviors associated with calving, including feeding, ruminating (while lying and standing), lying and standing in the days preceding calving, were quantified by an IMU sensor and classified automatically using LSTM-RNN models. In previous research, the multiple behaviors that have been identified as predictors of calving, were detected on an individual basis by the sensors. Accelerometers attached to a collar and an ear tag detected feeding and ruminating, while a

Conclusion

Seven behavior patterns associated with calving were identified and classified using an IMU sensor and an LSTM-RNN model. Feeding, ruminating (lying) and ruminating (standing), standing and lying were successfully detected and quantified on an individual basis in the days preceding and on the day of calving. This enables standing and lying behavior on the day of calving to be directly classified. Automatic classification and simultaneous measurement of each of these seven behaviors offers an

CRediT authorship contribution statement

Yingqi Peng: Conceptualization, Methodology, Software, Formal analysis, Writing - original draft, Investigation, Data curation. Naoshi Kondo: Writing - review & editing, Supervision, Funding acquisition. Tateshi Fujiura: Writing - review & editing, Project administration. Tetsuhito Suzuki: Writing - review & editing. Samuel Ouma: Investigation, Writing - review & editing. Wulandari: Software, Writing - review & editing. Hidetsugu Yoshioka: Investigation, Resources. Erina Itoyama: Investigation,

Declaration of Competing Interest

We declare that all authors have NO affiliations with or involvement in any organization or entity with any financial interest (such as honoraria; educational grants; participation in speakers’ bureaus; membership, employment, consultancies, stock ownership, or other equity interest; and expert testimony or patent-licensing arrangements), or non-financial interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed in

Acknowledgements

This study was supported by JSPS KAKENHI [Grant Number JP26252044]; We gratefully acknowledge the financial support of Ministry of Education, Culture, Sports, Science and Technology. We also want to thank the staff of Livestock Farm of Kyoto University for their help and support. Finally, I want to thank the financial support from the program of China Scholarships Council [No. 201508210203].

Cited by (38)

  • Classification of cow behavior patterns using inertial measurement units and a fully convolutional network model

    2023, Journal of Dairy Science
    Citation Excerpt :

    The data analysis process, from preprocessing to behavior pattern classification, is shown in Figure 3. Preprocessing and data annotation were similar to the methods described by Peng et al. (2019, 2020). In preprocessing, erroneous IMU data caused by the occasional data packet loss during the use of a micro-SD card were manually removed.

View all citing articles on Scopus
View full text