In beef cattle production, determining the ideal time to sell makes all the difference in a producer’s costs. This is because, once an animal reaches its natural growth limit, it stops gaining weight significantly. From that point on, every additional day in feedlot means spending money on feed without gaining meat in return. To identify this optimal window without having to weigh every animal in the herd on a scale – a time-consuming and stressful process for the cattle – a group of researchers supported by FAPESP has developed a noninvasive solution. They created a system that combines drones and artificial intelligence to map the growth of Nelore cattle, allowing for precise monitoring of fattening and slaughter decisions based on aerial images.
The technology was described in an article published in the journal Computers and Electronics in Agriculture . The details were presented by Jayme Garcia Arnal Barbedo , a researcher at Embrapa Digital Agriculture – one of the decentralized units of the Brazilian Agricultural Research Corporation (EMBRAPA) – in a presentation given on October 28 during FAPESP Week Netherlands .
“We’re very close to obtaining a complete, functional prototype of the system. The next crucial step will be to establish partnerships with companies in the sector or agtechs [ startups in the agricultural sector ] to transform that research into a commercially available product for ranchers,” Barbedo told Agência FAPESP .
The innovation uses aerial images captured by drones flying at a height of 15 meters. Using advanced deep learning algorithms, the system segments the animals’ bodies and extracts morphological metrics to map the fattening curve in real time. The results demonstrated that the optimal time for sale can be determined up to a week earlier than predicted by conventional methods.
“The weight gain curve of a group of animals intended for slaughter has a characteristic shape of an ‘S’: it starts slowly during the adaptation phase, enters a period of very accelerated gain, and then slows down in the final stage. From that point on, no matter how much feed you provide, the animal will no longer convert feed into meat at the same rate,” Barbedo explained. “Sometimes, a one-day difference in the decision to slaughter or in the sale of the herd can increase that operation’s profit by 5% to 10%,” he emphasized.
Eye on the sky
The system’s computational architecture combines two state-of-the-art artificial intelligence models. The first, for object detection, was fine-tuned to individually locate each Nelore steer in the feedlot, achieving 95.56% accuracy. The second model, meanwhile, allows for the precise delineation of the animal’s body contours at the pixel level.
Based on this information, the artificial intelligence performs a simple calculation: it compares the steer’s width to its length. The major advantage of this ratio is that it remains unaffected if the drone flies slightly higher or lower. As the animal gains muscle and fat in the feedlot, its body becomes visibly wider relative to its length. By observing how this measurement changes over the weeks, the system can accurately track the cattle’s weight gain without relying on the absolute size of the image in the photo.
To ensure data reliability in complex commercial environments, the researchers developed a filter. “The animals constantly move, lie down, or assume curved positions. Therefore, the artificial intelligence analyzes the entire batch and automatically discards individuals lying down or at unfavorable angles, retaining only animals standing and in a straight posture within the central window of the image. That allows us to calculate an extremely accurate average for the herd,” said Barbedo.
According to the researcher, the traditional method requires the herd to be periodically herded into pens to pass through the so-called “restraint chute.” This structure is a kind of narrow corridor designed to squeeze and immobilize the cattle, one by one, allowing them to be weighed on a physical scale. Forcing animals into this cramped space not only requires a great deal of work from the team but also increases stress and the risk of injuries and bruises in the cattle, in addition to causing frequent damage to the equipment itself. Because it is such an invasive and difficult operation, weigh-ins are usually performed only upon entry and exit from the feedlot (that is, on the first day of feedlot confinement and months later, upon shipment to the slaughterhouse), which ends up leaving producers “in the dark” for weeks.
The new system replaces physical stress with autonomous aerial observation. During field tests, analysts evaluated flights at altitudes of 5, 10, and 15 meters, finding that at 15 meters high, the cattle did not even notice the drone’s presence, continuing their natural grazing and resting behavior without any impact on animal welfare.
Validation under real-world conditions
The technology was tested and validated under real-world production conditions at Campanário Ranch, located in the municipality of Laguna Carapã in the state of Mato Grosso do Sul, a property that houses approximately 14,600 head of cattle in a feedlot system. The experiment monitored a herd of 110 Nelore cattle with an initial average age of 18 months over a 112-day fattening cycle, conducting 12 periodic aerial missions every ten days.
The collected data fed into a mathematical model capable of clearly identifying the three biological phases of development: the adaptation phase, the rapid growth phase, and the economic saturation plateau (when the animal reaches its body weight limit).
Upon detecting that the herd’s weight gain had entered the deceleration phase one week before the date projected by conventional management practices, the system indicated the optimal slaughter window in advance, thereby preventing unnecessary feed consumption. In the test batch of 110 animals, the resulting savings amounted to BRL 15,000. When scaled to the total feedlot capacity of the ranch (14,600 animals), the gains in operational and financial efficiency could reach approximately BRL 2 million.
The research was conducted under the auspices of the Science Center for Development in Digital Agriculture ( Semear Digital ), which is supported by FAPESP. It also received funding through four other projects ( 23/03870-8 , 24/15430-5 , 25/05985-2 , and 25/10602-5 ).
Other applications
In addition to monitoring the weight gain of beef cattle, the technology is being adapted to remotely identify stress-related behaviors in feedlots, such as disputes over food and misalignment at the feed trough (when images show that the animals are unable to line up side by side to eat, indicating a lack of space or pushing during feeding).
Another potential application is in dairy farming. Prompted by demands from the industry and international partners, EMBRAPA researchers are studying how to apply the system for the early detection of diseases such as mastitis and the analysis of dairy cows’ feeding patterns.
“The next step involves technology transfer to the market,” said Barbedo.