By capturing brain waves, the BrainTech technology can identify the judgment and classification a person makes while observing an image. Then, by simulating this process, the system can immediately and automatically label the image. The equipment helped to identify healthy leaves and leaves with powdery mildew or Asian soybean rust with high accuracy. The technology has several applications, such as the early identification of diseases in plantations or finding the most suitable pastures to maximize dairy production. The recognition system can be built in agricultural equipment, drones or cell phones. The same technology is used at airports to identify dangerous objects in suitcases. The equipment that allows the capture and simulation of brain waves started to be tested in Brazil in 2022 to detect early-stage diseases in soybean crops through artificial intelligence (AI). The work is made from a partnership between Embrapa and the companies Macnica DHW and InnerEye; the latter developed BrainTech, an equipment that captures the neural signals of experts through a helmet with electrodes, in a process that resembles an electroencephalogram (EEG).The system then simulates how the brain works when the experts see images of diseased plants, automating and making the labeling stage faster and more efficient. With that, the researchers hope to accelerate decision-making, and thus reduce losses in rural enterprises and rationalize the use of natural resources. “This is a pioneering initiative by Embrapa that has the disruptive BrainTech technology, brought exclusively by Macnica DHW to Brazil. By associating EEG neural signals and AI, it is possible to create a machine that mimics the human brain with high reliability,” observes Macnica DHW IoT & AI Solutions manager Fabrício Petrassem. The system's testing and validation had the participation of developer Yonatan Meir, from InnerEye, who came from Israel in August especially for this purpose. “By capturing brain waves, InnerEye's solution can identify a person's judgment and classification of an image, allowing that image to be automatically and immediately labeled,” Meir explains. The system is already used at European airports to identify dangerous objects in suitcases. In 2019, Macnica DHW sought Embrapa as a partner to explore the technology in the agricultural sector, with new possible applications. The first was the early detection of diseases in plants, whose experiments began in April 2022. The researcher at Embrapa Digital Agriculture Jayme Barbedo explains the operation of technology that uses artificial intelligence applied to agriculture The experiment “AI tools have evolved a lot and, with good quality data, they can solve almost any problem,” states Embrapa Agricultura Digital researcher Jayme Barbedo, who leads the project on the Embrapa's side. The challenge, according to him, is to obtain such ‘quality data’, which needs to be not only collected but also labeled by experts. A costly and time-consuming process in which the equipment will help. The first results of the experiment were positive, as the equipment helped to identify, with high accuracy, (powdery mildew and soybean rust) infected leaves and healthy ones. Now the project will go beyond the detection of diseased/non-diseased plants and advance in identifying the type of disease present in the soybean plantation, starting with the most commercially significant ones. The inclusion of corn and coffee crops in the experiments is also being negotiated with the respective Embrapa research centers. In April, the equipment was brought to Brazil to the headquarters of Macnica DHW, a Japanese multinational located in Florianópolis, SC. There they assembled the structure for the experiment to capture the brain waves of plant pathologists Cláudia Godoy and Rafael Soares (pictured on the left) from Embrapa Soybean. Both evaluated about 1,500 images of diseased and healthy leaves in tests with the collector helmet. The proof-of-concept stage showed that the models generated from the experts' electroencephalograms can handle images well, which allowing to train the machine to identify diseased plants. “The combination of the labeled images – diseased/healthy – with the experts' brain signals resulted in improved model performance, pointing to the feasibility of using AI”, Barbedo points out. Artificial Intelligence Research area that aims to design, develop, apply and assess methods and techniques to create intelligent systems that can acquire and integrate knowledge about the domain in which they operate on their own, in order to progressively improve their performance to achieve their goals. First impressions “The experiment was very interesting, as the system learns to identify images of sick leaves through a count performed silently by identifying brain waves when one views the pictures of diseased and healthy leaves, which are quickly shown on a computer screen,” reports Cláudia Godoy (pictured above). “With the evolution of artificial training, such recognition technologies can be used by people who do not have much knowledge of diseases and help management”, she details. According to Soares, two diseases were chosen for this experiment: Asian rust, the most economically important disease that affects the crop, and powdery mildew, which is relevant in the Brazilian South. “These diseases were chosen not only due to the impact they generate for soybean cultivation, but also because they cause two distinct types of leaf symptoms in the plant; in addition, there was a a suitable availability of images for assessment,” Soares explains. For the researcher, the improvement of soybean disease management tools is relevant because “detecting and diagnosing diseases is one of the largest difficulties in crop management, and innovative technologies that add information to such practices are desirable and necessary”, he points out. Asian rust costs over US$2 billion per harvest in Brazil Since its introduction in Brazil in 2001, Asian soybean rust (picture on the right), caused by the fungus Phakopsora pachyrhizi, has been the most severe crop disease, and can lead to losses of up to 80% if left uncontrolled. According to surveys by the Anti-Rust Consortium, costs with the disease exceed US$2 billion per harvest in Brazil, considering the acquisition of fungicides and the productivity losses it causes. Management strategies are focused on practices such as the sanitary break, which is the period of at least 90 days the fields are left without live soybean plants to reduce the fungal infestation. Control strategies also include the use of early cycle cultivars and sowing at the beginning of the recommended crop season, the adoption of resistant cultivars, compliance with the sowing calendar and the use of fungicides. The fungus P. pachyrhizi currently has mutations that give it resistance to the three main groups of site-specific fungicides, and new mutations can be selected over time. “The fungus that causes the disease can adapt to some of the control strategies, either by losing sensitivity to fungicides or by ‘breaking’ the genetic resistance of soybean cultivars,” Cláudia Godoy explains. Therefore, Embrapa's recommendation is for farmers to adopt the management strategies available to preserve the fungicides and cultivars available. “When used in combination, all of these strategies have allowed an adequate management of the disease,” Godoy asserts. Powdery mildew can cause losses of up to 35% Although powdery mildew does not have the same economic impact as Asian soybean rust, there are reports of productivity losses ranging from 10% to 35%. The disease is caused by the fungus Erysiphe diffusa, which causes a thin whitish cover either in small patches or throughout the entire aerial portion of the plant, especially the leaves. In severe infections, the leaves may dry out and fall prematurely. The disease is favored by periods of low humidity and mild temperatures (18°C to 24°C), and it is more common in the southern region of Brazil, at higher altitudes with late sowing, due to the greater climate favorability. “Disease control strategies involve the use of resistant cultivars and chemical control,” the researcher Rafael Soares informs. Photos (rust and mildew): Embrapa How the BrainTech technology works Image: InnerEye The system “mimics” how the experts' brains operate when they visualize images of diseased plants, automating and making the labeling stage faster and more efficient. The idea is to as closely as possible simulate the brain process of a specialist when they identify something or make a decision, as was done with the plant pathologists. The first step is to calibrate the model by adjusting a helmet with electrodes on an expert's head to identify their brain waves. “Each person has a different brain pattern, that is, brain electrical signals are different from person to person. Therefore, it is necessary to perform calibration for each person so that the model understands what they are thinking,” Barbedo explains. Once the system has ‘learned’ how the person works, the process of labeling for the database begins. The experts are instructed to number (1, 2, 3 …) the sick leaves when they see them on the screen, which shows three images per second. The system captures the brain waves emitted with each new stimulus, which are different when a healthy leaf is seen. According to the project leader, the counting process is not mandatory, but it reinforces brain signals, making it easier to differentiate between diseased and healthy. The system allows the viewing of up to ten images per second. Reliability of results With an average length of half an hour, in each session it was possible to label over a thousand images, a task that would take days in a manual system. In addition to the gain in speed in the labeling process, Barbedo underscores the reliability of the system, “which has mechanisms to correct possible errors, making the trained model more reliable.” Through neural signals, the system can identify whether the specialist blinked or lost attention in the process of viewing the sequence of images . In those cases, the system discards the result and reintroduces the image later. The BrainTech system generates an indicative attention curve, and it pauses the experiment for a break when it falls to a critical level for the reliability of the results. In addition, the system can detect the expert's level of certainty upon viewing the image, which is called a soft label. The use of this parameter allows for a better calibration of the model according to the level of experience of each expert; consequently, this brings higher accuracy in the AI model's decision. Applications in agriculture The technology opens up several possibilities of application in the agricultural sector. The trained models could be embedded into agricultural equipment or cell phone apps and operate in areas where there is a shortage of specialized labor. The more rational application of pesticides, with less economic costs and less environmental impact, and a cleaner and more sustainable food production would be possible with trained models built in machinery, which would identify the need to apply pesticides as they go through production lines, in real time and in specific parcels. “Embedding this model in an app would expedite farmers' decision-making when diseases and symptoms of pathologies are identified, accelerating the adoption of the necessary measures,” Barbedo asserts. The researcher also points out the relevance of the use of the technology in the rotation of pastures for dairy cattle, an area where there is a lack of specialists. The choice of the most appropriate parcels to maximize milk production is made by a technician who is experienced in identifying the best location and the ideal amount of animals. “The system could simulate such specialist's action to assert a location. Most properties don't have someone with that expertise,” he concludes. The partners InnerEye a section of the BrainTech company (partner of the Macnica Inc. group) from Israel that combines artificial intelligence (AI) with human intelligence, through the capture and interpretation of brain waves. Macnica DHW is the South American arm of the Japanese group Macnica Inc., a distributor of semiconductors and supplier of Internet of Things (IoT) and AI solutions. It operates in areas such as cybersecurity, autonomous vehicles, robotics, AI, industrial internet of things (IIoT), commercial/banking automation. It develops products and solutions ranging from custom hardware to the development of artificial intelligence algorithms.