Zobraziť v EDS

Automated detection of broiler chicken behaviors through the integration of 3D accelerometer sensor and machine learning techniques.

Uložené v:
Podrobná bibliografia
Názov: Automated detection of broiler chicken behaviors through the integration of 3D accelerometer sensor and machine learning techniques.
Autori: Sakib, Mohammad1 (AUTHOR) sakib.eee@ulab.edu.bd, Alam, S. K. Shah1 (AUTHOR), Atul, S. H. Mridha1 (AUTHOR), Supto, F. H.1 (AUTHOR), Hossain, M. Mofazzal1 (AUTHOR) mofazzal.hossain@ulab.edu.bd
Zdroj: AIP Conference Proceedings. 2024, Vol. 3242 Issue 1, p1-19. 19p.
Predmety: *MACHINE learning, *POULTRY farm management, *SUPPORT vector machines, *K-nearest neighbor classification, *ANIMAL behavior, *POULTRY farms
Abstrakt: Broiler behavior recognition systems, utilizing neck-mounted accelerometers, offer a non-invasive and cost-effective solution for assessing broiler conditions. These systems hold great potential for improving poultry welfare and farm management. However, the complexity of animal behavior poses challenges, as similar acceleration data can represent different behaviors. Nevertheless, prior investigations in this field have encountered limitations stemming from the utilization of diminutive datasets and brief sampling durations, thereby resulting in restricted generalizability. This study presents an efficient system for recognizing six broiler behaviors: sitting, standing, feeding, drinking, walking, and preening, based on 30 broilers aged 7±1-weeks observed over 24h at a 40 Hz sampling frequency. The data were meticulously annotated and categorized into six distinct labels. Machine learning algorithms including Decision Trees (DT), Discriminant Analysis (DA), K-Nearest Neighbor (KNN), Support Vector Machines (SVM), and Ensemble were employed for classification. The average performance metrics for these algorithms, including accuracy, sensitivity, specificity, precision, recall, and F-1 score, were as follows: (DT = 0.8429, 0.8428, 0.8370, 0.8296, 0.8238, 0.8530), (DA = 0.7148, 0.7194, 0.7147, 0.7121, 0.7120, 0.7232), (KNN = 0.8333, 0.8386, 0.8344, 0.8270, 0.8420, 0.8429), (SVM = 0.8754, 0.8796, 0.8774, 0.8692, 0.8677, 0.8840), and (Ensemble = 0.8408, 0.8447, 0.8319, 0.8244, 0.8206, 0.8503). This method successfully and precisely detected particular broiler behaviors, indicating its potential for use in real-world settings in the future. [ABSTRACT FROM AUTHOR]
Databáza: Academic Search Index
Popis
Abstrakt:Broiler behavior recognition systems, utilizing neck-mounted accelerometers, offer a non-invasive and cost-effective solution for assessing broiler conditions. These systems hold great potential for improving poultry welfare and farm management. However, the complexity of animal behavior poses challenges, as similar acceleration data can represent different behaviors. Nevertheless, prior investigations in this field have encountered limitations stemming from the utilization of diminutive datasets and brief sampling durations, thereby resulting in restricted generalizability. This study presents an efficient system for recognizing six broiler behaviors: sitting, standing, feeding, drinking, walking, and preening, based on 30 broilers aged 7±1-weeks observed over 24h at a 40 Hz sampling frequency. The data were meticulously annotated and categorized into six distinct labels. Machine learning algorithms including Decision Trees (DT), Discriminant Analysis (DA), K-Nearest Neighbor (KNN), Support Vector Machines (SVM), and Ensemble were employed for classification. The average performance metrics for these algorithms, including accuracy, sensitivity, specificity, precision, recall, and F-1 score, were as follows: (DT = 0.8429, 0.8428, 0.8370, 0.8296, 0.8238, 0.8530), (DA = 0.7148, 0.7194, 0.7147, 0.7121, 0.7120, 0.7232), (KNN = 0.8333, 0.8386, 0.8344, 0.8270, 0.8420, 0.8429), (SVM = 0.8754, 0.8796, 0.8774, 0.8692, 0.8677, 0.8840), and (Ensemble = 0.8408, 0.8447, 0.8319, 0.8244, 0.8206, 0.8503). This method successfully and precisely detected particular broiler behaviors, indicating its potential for use in real-world settings in the future. [ABSTRACT FROM AUTHOR]
ISSN:0094243X
DOI:10.1063/5.0231888