Automatic Newcastle disease detection using sound technology and deep learning method
Introduction
Animal welfare is a topic of concern to the poultry industry and the public, and it is closely related to the quality of poultry products and the health of consumers. Good animal welfare can promote the healthy growth of poultry, reduce the case fatality rate, and improve quality and productivity (Li et al., 2020, Rowe et al., 2019, Herborn et al., 2020). In modern poultry production, disease threatens the health of poultry every year, which not only seriously affects the animal welfare of poultry but also directly results in serious economic losses. Newcastle disease (ND) is one of the most widespread and harmful diseases in poultry. It causes respiratory and nervous system diseases and has a negative impact on the survival and breeding of poultry (Alexander, 2001, Ganar et al., 2014).
Early warning of poultry diseases is very important for poultry health. Automated disease detection systems are of great significance to poultry production and animal welfare (Neethirajan et al., 2017). With the development of precision livestock farming (PLF), increasingly more studies have used image and sound technology to monitor animals. PLF methods collect the images and sounds of animals in a noncontact way, which can be used for monitoring individual and group animals and assessing environmental and animal welfare (Aydin, 2017, Bishop et al., 2019, Zhuang et al., 2018, Fang et al., 2020). Compared with other technologies, sound technology has significant advantages. As noncontact information collection equipment, microphones are inexpensive, widely used, and not limited by sight and light. A large group of animals can be monitored with one sensor (Chelotti et al., 2016, Carpentier et al., 2018, Mcloughlin et al., 2019).
Vocalization is an important indicator to evaluate poultry welfare, and it can reflect the feeding, growth and health of poultry, among other information (Fontana et al., 2016, Du et al., 2020, Zong et al., 2021). For example, Lee et al. (2015) used three binary-classifier support vector machines (SVMs) to detect the stress from the changes in the sounds of poultry and classified the changes into subsidiary sound types. The accuracy of the proposed method in detecting stress approached 96.2%. Jakovljević et al. (2019) used audio features and an SVM to classify poultry calls at different temperature states, with an accuracy of 63% to 83% for different age groups. Aydin et al. (2016) monitored the short-term feeding behaviour of 10 broilers, investigated and found the relationship between the feeding behaviour obtained by the algorithm and the feeding behaviour recorded by a weighing scale and video camera, and found a strong positive correlation between these methods. Sadeghi and Banakar (2017) proposed an intelligent fowl sexing system based on data mining methods to distinguish hens from cock hatchlings, used improved distance evaluation (IDE) to select the best features, and classified fowl sounds using an SVM.
Some scholars evaluated poultry health by detecting coughs, sneezes and rales (Carroll et al., 2014, Rizwan et al., 2016, Carpentier et al., 2019, Liu et al., 2020) while others evaluated poultry health using poultry vocalization (Huang et al., 2019, Mahdavian et al., 2020, Cuan et al., 2020). Sadeghi et al. (2015) proposed an intelligent method for the detection and classification of chickens infected by Clostridium perfringens type A based on their vocalization. Banakar et al. (2016) used 25 statistical features from frequency domains and 75 statistical features from time-frequency domains to diagnose avian diseases by vocalization. Alex et al. (2018) used audio feature extraction with the k-nearest neighbors method machine learning technique to detect avian influenza calls. Mahdavian et al. (2021) studied the difference in the call signals of infectious bronchitis and Newcastle disease. Five different acoustic features and an SVM were used to determine bird health conditions, and the results showed that the MFCC had better performance in detecting Newcastle disease birds. Deep learning is one of the most important methods in current scientific research. It has been used in audio research on animals, livestock and poultry and has shown good results (Küc̣üktopcu et al., 2019, LeBien et al., 2020, Nunes et al., 2021, Maegawa et al., 2021). However, there are few studies using deep learning methods to detect sick poultry calls.
In addition, in most studies of poultry sound, poultry vocalizations are extracted artificially or selected from smaller noise audio data by the threshold value method. However, the real breeding environment contains considerable complex noise, such as fans, air conditioners, and human and poultry behaviour noises, so it is difficult to obtain poultry vocalizations using previous methods. The objective of this study is to develop a new deep poultry vocalization network (DPVN) method based on audio technology and deep learning to detect Newcastle disease. The poultry vocalizations were automatically extracted from complex noise. Furthermore, the bidirectional long short-term memory neural network was used to recognize the vocalizations of poultry infected with Newcastle disease.
Section snippets
Experimental setup
The poultry used in the experiment were specific-pathogen-free (SPF) chickens, as shown in Fig. 1(a). The experiment was conducted in the poultry isolation cage of the animal safety laboratory at South China Agricultural University to ensure that the normal activities of poultry would not be disturbed by other viruses during the experiment. Twenty 15-day-old SPF chickens were placed in poultry isolation cages a week early to adapt to the environment. After one week of feeding, they were
Vocalization endpoint detection
The endpoints of the poultry vocalizations in the three random 20-minute audio files were detected using the multiple subband poultry vocalization endpoint detection method. These endpoints were labelled by professionals artificially, and a total of 4372 poultry vocalizations were labelled. The endpoint detection results are shown in Table 3. A total of 4307 vocalizations were detected in three audio samples, 4158 of which were correctly detected. The remaining 149 vocalizations were unlabelled
Conclusion
Combined with sound technology and deep learning methods, a new method for the early detection of Newcastle disease was designed in this paper. Different deep learning methods were compared in the experiments, and the performance of the methods was evaluated using different statistical indices. The results of this study showed that the proposed method can be used to detect the vocalizations of poultry infected with Newcastle disease in the early stage. After comparing the results of five
CRediT authorship contribution statement
Kaixuan Cuan: Conceptualization, Methodology, Software, Writing – original draft. Tiemin Zhang: Supervision, Conceptualization. Zeying Li: Validation, Writing – review & editing. Junduan Huang: Validation, Methodology. Yangbao Ding: Data curation. Cheng Fang: Visualization, Software.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
Funding: This work was supported by the Guangdong HUST Industrial Technology Research Institute, Guangdong Provincial Key Laboratory Of Digital Manufacturing Equipment [grant No. 2020B1212060014]; Chaozhou Science and Technology project [grant No. 202101ZD07]; Lingnan Modern Agricultural Science and Technology Guangdong Provincial Laboratory Maoming Laboratory independent scientific research project [grant No. 2021ZZ003] and the Guangdong Province Special Fund for Modern Agricultural Industry
References (42)
- et al.
Using sound technology to automatically detect the short-term feeding behaviours of broiler chickens
Comput. Electron. Agric.
(2016) Using 3D vision camera system to automatically assess the level of inactivity in broiler chickens
Comput. Electron. Agric.
(2017)- et al.
An intelligent device for diagnosing avian diseases: Newcastle, infectious bronchitis, avian influenza
Comput. Electron. Agric.
(2016) - et al.
Livestock vocalisation classification in farm soundscapes
Comput. Electron. Agric.
(2019) - et al.
Automatic cough detection for bovine respiratory disease in a calf house
Biosyst. Eng.
(2018) - et al.
Development of sound-based poultry health monitoring tool for automated sneeze detection
Comput. Electron. Agric.
(2019) - et al.
A real-time algorithm for acoustic monitoring of ingestive behavior of grazing cattle
Comput. Electron. Agric.
(2016) - et al.
Detection of avian influenza-infected chickens based on a chicken sound convolutional neural network
Comput. Electron. Agric.
(2020) - et al.
Comparative study on poultry target tracking algorithms based on a deep regression network
Biosyst. Eng.
(2020) - et al.
Vocalisation sound pattern identifification in young broiler chickens
Animal
(2016)
Newcastle disease virus: Current status and our understanding
Virus Res.
Method for detecting avian influenza disease of chickens based on sound analysis
Biosyst. Eng.
An intelligent method for detecting poultry eating behaviour based on vocalization signals
Comput. Electron. Agric.
A real-time bird sound recognition system using a low-cost microcontroller
Appl. Acoust.
A pipeline for identification of bird and frog species in tropical soundscape recordings using a convolutional neural network
Ecol. Inf.
Review: Automated techniques for monitoring the behaviour and welfare of broilers and laying hens: towards the goal of precision livestock farming
Animal
A new survey method using convolutional neural networks for automatic classification of bird calls
Ecol. Inf.
Acoustic features of vocalization signal in poultry health monitoring
Appl. Acoust.
Ability evaluation of a voice activity detection algorithm in bioacoustics: A case study on poultry calls
Comput. Electron. Agric.
Recent advancement in biosensors technology for animal and livestock health management
Biosens. Bioelectron.
Horse foraging behavior detection using sound recognition techniques and artificial intelligence
Comput. Electron. Agric.
Cited by (23)
Research on broiler health status recognition method based on improved YOLOv5
2023, Smart Agricultural TechnologyNon-contact sensing technology enables precision livestock farming in smart farms
2023, Computers and Electronics in AgricultureInternet of Things and Machine Learning techniques in poultry health and welfare management: A systematic literature review
2022, Computers and Electronics in AgricultureCitation Excerpt :Features commonly used in previous studies (i.e., Aydin et al., 2015; Aydin & Berckmans, 2016; Banakar et al., 2016; Fontana et al., 2015) for detecting and classifying poultry vocalization data include pitch, frequency, and time–frequency. Similarly, studies such as those from Carroll et al. (2014), Cuan et al. (2022b), Cuan et al. (2022a) and Jung et al. (2021) used the Mel frequency cepstrum coefficients (MFCCs) to represent the acoustic sound of hens to recognize behavioral meanings from those sounds. This subsection briefly describes data pre-processing techniques for poultry audio and image data.
Gender determination of domestic chicks based on vocalization signals
2022, Computers and Electronics in AgricultureCitation Excerpt :A large number of vocalizations can obtain more representative results. Deep learning has shown good performance in the detection of calls of different animals (Zhong et al., 2021; Nunes et al., 2021; Cuan et al., 2022), but there is a lack of deep learning methods in gender detection of chick vocalizations. This study uses three deep learning models, CNN, LSTM, and GRU, to detect the sex of chick vocalizations, and the test accuracies are 74.55%, 75.73%, and 76.15%, respectively.