RESUMO
The study aimed to forecast ammonia exposure risk in broiler chicken production, correlating it with health injuries using machine learning. Two chicken breeds, fast-growing (Ross®) and slow-growing (Hubbard®), were compared at different densities. Slow-growing birds had a constant density of 32 kg m-2, while fast-growing birds had low (16 kg m-2) and high (32 kg m-2) densities. Initial feeding was uniform, but nutritional demands led to varied diets later. Environmental data underwent selection, pre-processing, transformation, mining, analysis, and interpretation. Classification algorithms (decision tree, SMO, Naive Bayes, and Multilayer Perceptron) were employed for predicting ammonia risk (10-14 pmm, Moderate risk). Cross-validation was used for model parameterization. The Spearman correlation coefficient assessed the link between predicted ammonia risk and health injuries, such as pododermatitis, vision/affected, and mucosal injuries. These injuries encompassed trachea, bronchi, lungs, eyes, paws, and other issues. The Multilayer Perceptron model emerged as the best predictor, exceeding 98% accuracy in forecasting injuries caused by ammonia. The correlation coefficient demonstrated a strong association between elevated ammonia risks and chicken injuries. Birds exposed to higher ammonia concentrations exhibited a more robust correlation. In conclusion, the study effectively used machine learning to predict ammonia exposure risk and correlated it with health injuries in broiler chickens. The Multilayer Perceptron model demonstrated superior accuracy in forecasting injuries related to ammonia (10-14 pmm, Moderate risk). The findings underscored the significant association between increased ammonia exposure risks and the incidence of health injuries in broiler chicken production, shedding light on the importance of managing ammonia levels for bird welfare.
RESUMO
Vocalization seems to be a viable source of signal for assessing broiler welfare. However, it may require an understanding of the birds' signals, both quantitatively and qualitatively. The delivery of calls with a specific set of acoustic features must be understood to assess the broiler's well-being. The present study aimed to analyze broiler chick vocalization through the sounds emitted during social isolation and understand what would be the flock size where the chicks present the smallest energy loss in vocalizing. The experiments were carried out during the first 3 days of growth, and during the trial, chicks received feed and water ad libitum. A total of 30 1-day-old chicks Cobb® breed were acquired at a commercial hatching unit. The birds were tested from 1 to 3 days old. A semi-anechoic chamber was used to record the vocalization with a unidirectional microphone connected to a digital recorder. We placed a group of 15 randomly chosen chicks inside the chamber and recorded the peeping sound, and the assessment was conducted four times with randomly chosen birds. We recorded the vocalization for 2 min and removed the birds sequentially stepwise until only one bird was left inside the semi-anechoic chamber. Each audio signal recorded during the 40 s was chosen randomly for signal extraction and analysis. Fast Fourier transform (FFT) was used to extract the acoustic features and the energy emitted during the vocalization. Using data mining, we compared three classification models to predict the rearing condition (classes distress and normal). The results show that birds' vocalization differed when isolated and in a group. Results also indicate that the energy spent in vocalizing varies depending on the size of the flock. When isolated, the chicks emit a high-intensity sound, "alarm call", which uses high energy. In contrast, they spent less energy when flocked in a group, indicating good well-being when the flock was 15 chicks. The weight of birds influenced the amount of signal energy. We also found that the most effective classifier model was the Random Forest, with an accuracy of 85.71%, kappa of 0.73, and cross-entropy of 0.2.
RESUMO
Feeding is one of the most critical processes in the broiler production cycle. A feeder can collect data of force signals and continuously transform it into information about birds' feed intake and quickly permit more agile and more precise decision-making concerning the broiler farm's production process. A smart feeding unit (SFU) prototype was developed to evaluate the broiler pecking force and average feed intake per pecking (g/min). The prototype consisted of a power supply unit with a data acquisition module, management software connected to a computer for data storage, and a video camera to verify the pecking force during signal processing. In the present study, seven male Cobb-500 broilers were raised in an experimental chamber to test and commission the prototype. The prototype consisted of a feeding unit (feeder) with a data acquisition module (amplifier), with real-time integration for testing and intuitive operation with Catman Easy software connected to a computer to obtain and store data from signals. The sampling of average feed intake per pecking per broiler (g) was conducted during the first minute of feeding, subtracting the amount of feed provided per the amount of feed consumed, including the count of pecking in the first minute of feeding. An equation was used for estimating the average feed intake per pecking per broiler (g). The results showed that the average broiler pecking force was 1.39 N, with a minimum value of 0.04 N and a maximum value of 7.29 N. The average feed intake per pecking (FIP) was 0.13 g, with an average of 173 peckings per minute. The acquisition, processing, and classification of signals in the pecking force information were valuable during broilers' feeding. The smart feeding unit prototype for broilers was efficient in the continuous assessment of feed intake and can generate information for estimating broiler performance.