Indeed, “Transformation in Livestock farming” was decided to be the main theme of the 2022 edition of the world’s leading trade fair for animal farming and livestock industry, EuroTier in Hannover. Exhibitors predict that social and ecological requirements as well as animal, meat and dairy production chains in Europe have to evolve in the spirit of the EU initiative “Green Deal”. Creating a sustainable food system is one of the key goals of this EU programme. And for food production, the most important challenges are climate neutrality and animal friendliness. The breeding processes should then become more efficient at both smaller and larger scale farms and ensure animal welfare.
Such optimization can be achieved with the use of precision and smart farming technologies that can collect and process huge amounts of data to support decisions and daily routines at the farms. The technology enables breeders to collect information about animals and interpret this data. Sensors can monitor many animal health parameters, such as temperature or heart rate. Also, information about behavior patterns can be collected and then analyzed using machine learning algorithms. Based on this information, the farmer can make optimal changes to animal care.
One of the pillars of precision farming systems are computer vision and object detection techniques. The neural networks can be trained to process huge amounts of visual data and analyze images to provide actionable insights in real time, non-intrusive manner. Exemplary applications include monitoring of livestock’s health, food and water supply control, detecting abnormal behavior and livestock counting. All of these applications require a dedicated surveillance system (often equipped with high definition cameras) that can monitor the livestock at the farm or pastures. Vast open areas can be also effectively monitored by drones, which can collect information from the air providing a clear top-down view of the supervised region.
Precision and smart farming aim to ensure the most sustainable food production process. For example, the system can continuously monitor the animals on the pasture and analyze their behavior, ambient temperature value or water level in the tank. Based on the sensors’ readings and video stream enhanced with computer vision analytics, the system detects whether the animals need more water or food. The farmers get informed on the situation in real time so that they can react properly in time. Also, the collected data can be used for forecasting to support farmer’s decisions and planning.
Another application of computer vision systems in precision farming is planning the passage of agricultural machinery based on the satellite images. The route can be chosen in terms of the current needs and state (eg. mowing) and reduce the related costs.
Another area that adapted this kind of innovation is aquaculture. In precision aquaculture, the technology uses a variety of interconnected sensors to monitor conditions on fish farms. Synthesized data help farm operators take decisions that improve farmed fish health and farm profitability. At the same time, these solutions allow to minimize the impact of farming on the environment.
Health inspection is one of the basic applications of computer vision in smart livestock and poultry farming. The captured images are processed to detect changes in the animal behavior or appearance and recognize early symptoms of an infection. Also any registered fluctuations in appetite or apathy of animals may indicate a threat of developing disease. Early detection of the infectious disease occurrences enables targeted action that can prevent the disease from spreading and avoid loss in the livestock. Smart farming systems allow to quickly detect deviations from the norm in the behavior of cattle or pigs and implement effective prevention.
Computer vision systems have proved to be effective in health inspection also in other specific, hardly accessible environments. One example is the honeybee colony inspection. The video feed from the camera installed in the hive is analyzed to recognize unusual honeybee appearance or behavior and alert the beekeepers on the anomalies.
.
Annotating data for computer vision in agriculture involves labeling and marking specific objects or regions of interest within images to train machine learning models.
First define annotation types, determine the specific objects or features you want to annotate in your agricultural images. For example, you might want to label animals, fields, buildings, or specific disease symptoms. The next step is to gather a diverse set of images that cover various agricultural scenarios, including different object types, lighting conditions, and angles. Ensure that the images capture the objects or features you want to annotate.
Decide on type of annotation, e.g. bounding boxes are commonly used to annotate objects in computer vision, it is good for easy data, such as animals or fields. Remember to adjust the boxes to tightly encapsulate the objects while avoiding unnecessary inclusion of background. For more detailed annotations, especially when dealing with complex shapes or overlapping objects, you can use segmentation masks. Segment each object or region of interest pixel by pixel using tools that allow you to draw or paint within the boundaries. Don’t forget to Include class labels! Assign appropriate class labels to each annotation. For example, if you are annotating different crop types, assign labels such as “cow,” “horse,” or “farm” to each bounding box or segmentation mask.
