.JPG)
The Brazilian coffee chain should incorporate in the coming years an intelligent and unprecedented system to speed up and improve the beverage certification process. The new technology employs advanced computing techniques to automatically analyze and classify the overall quality of the coffee in minutes. The analysis is performed without the need for beverage preparation, as the technology examines images of the powder. Today, the most reliable quality assessment is the sensory analysis in which human tasters taste the drink.
Named CoffeeClass, the technology was developed by researchers from Embrapa Instrumentação (SP) and uses computer vision and artificial intelligence (AI) to measure global quality directly in roasted and ground coffee. The development and application of the system can contribute to the implementation of public policies and standards created for the sector, in addition to bringing consumers closer to quality certification systems. Coffee drives one of the main chains of Brazilian agribusiness, which generates an annual revenue of around US$ 5 billion for the country.
The technological solution interprets patterns in enlarged images of roasted and ground coffee and correlates them with its overall quality identified in the beverage. For this, it uses reflectance and fluorescence imaging techniques, which use different lights under the roasted and ground coffee sample to show compounds linked to the quality of the beverage.
In the future, consumers will use technology in the supermarket
The artificial intelligence built into the system learns to “look” at the sample, extract patterns and predict the global quality class of the coffee being analyzed. AI is a research area that brings together several disciplines and seeks to develop computational methods capable of imitating human cognitive skills, such as learning, logic and process optimization.
For now, the smart tool is a prototype, but it is already born with high potential to provoke innovation in the coffee value chain. It is believed that, in the future, the system may be used by producers in the field and even by consumers in supermarkets.
Smart computing to understand coffee
Research with CoffeeClass began in 2013 and was developed by Embrapa analyst Ednaldo José Ferreira. The work was carried out within the scope of the Research Café Consortium, created in 1997 and which involves about 50 teaching, research and rural extension institutions, under the coordination of Embrapa Café. Ferreira says that the research arose from a lack of the chain that demanded an automated tool for the analysis of coffee.
The proposal was to generate an automated tool to assist the work of coffee tasters – specialists in sensory analysis – and thus consolidate a technology specifically aimed at certification based on the Global Quality (QG) of the ABIC Coffee Quality Program (PQC). . The proof of concept and the prototype were initially based on the classification of this program, but, according to Ferreira, the technology can be adapted to other indicators of coffee quality.
The HQ is the joint perception of characteristics, such as aroma, flavor, body, acidity, astringency, bitterness, fragrance and others that contribute to the composition of an overall note within a sensory scale - 0 to 10 for the ABIC PQC - used as an indicator of product quality. It is performed by a team of trained specialists who use the standardized preparation of a coffee drink as the basis of the analysis protocol, a process popularly known as "cup proof".
The PQC, launched in 2004, uses this QG score to certify the product in three categories: “Gourmet”, for very high quality coffees; “Superior”, for relatively high quality coffees and “Traditional/Extra strong”, for coffees of recommendable quality and affordable price.
An easy-to-use technology
Doctor in Computer Science and Computational Mathematics, Ferreira explains that several tools were tested before arriving at CoffeeClass. “The focus of the research has always been to generate a simple, low-cost apparatus that could be operated without difficulties, with minimal commands, by all actors in the coffee chain, on the farm itself, before leaving the field, in laboratories, cooperatives, , roasters, supermarkets, coffee shops, among others”, he says.
Ferreira used Data Science to carry out the pilot study with 120 samples of roasted and ground coffee, from various regions of the country, already certified by the PQC, which made it possible to generate a database with more than 2,400 digital images. “We identified the potential and designed a module called the smart core, based on a set of Artificial Intelligence methods and techniques that aim to automatically improve the performance of the analysis”, he explains.
In laboratory tests, the technology identified the Gourmet, Superior, Traditional and Non-Recommended coffee categories with a percentage slightly higher than 75% hit for blind samples.
The study involved low-cost commercial digital microscopes that use different LEDs (light emitting diodes) as a source of illumination for a sample of approximately one gram, in addition to a sample holder and software developed by Embrapa Instrumentação. .
Ferreira explains that the interaction of different lights with the coffee sample produces patterns in images enlarged by digital microscopes, which make it possible to characterize the coffee. The lights at different wavelengths allow detecting the composition of the coffee, as well as mixtures with low quality beans (defects and others), which interfere with the overall quality.
“The color variations observed in enlarged digital images, acquired directly from a sample of roasted and ground coffee, reveal characteristics closely linked to the overall quality of the product”, he explains.
The tool showed significant success potential in laboratory analyses, however, the project leader believes that it is possible to go further and develop other indicators to improve CoffeeClass.
“The challenge now is to understand the characteristics revealed by the intelligent system and expand the horizon of application of the technology”, says Ferreira. To carry out this new stage of the research, he says that he prepared and submitted a new project to the Consórcio Pesquisa Café program, with which he intends to evolve the study.
In the next phase, the work will bring together diverse expertise from universities, research institutes, government agencies, such as the Ministry of Agriculture, Livestock and Supply (Mapa) and its official laboratories (Lanagros), in addition to other historical and recognized entities. excellence in the coffee chain.
“In order to automate and modernize these analyses, Embrapa Instrumentação, with the support of Consórcio Pesquisa Café and the partnership with ABIC, will finalize and launch this technology, unprecedented in the world. The CoffeeClass will undoubtedly revolutionize the quality classification of Brazilian Coffees, so that, more and more, we can consolidate and conquer new markets”, says the head of Technology Transfer at Embrapa Café, Lucas Tadeu Ferreira.
Named CoffeeClass, the technology was developed by researchers from Embrapa Instrumentação (SP) and uses computer vision and artificial intelligence (AI) to measure global quality directly in roasted and ground coffee. The development and application of the system can contribute to the implementation of public policies and standards created for the sector, in addition to bringing consumers closer to quality certification systems. Coffee drives one of the main chains of Brazilian agribusiness, which generates an annual revenue of around US$ 5 billion for the country.
The technological solution interprets patterns in enlarged images of roasted and ground coffee and correlates them with its overall quality identified in the beverage. For this, it uses reflectance and fluorescence imaging techniques, which use different lights under the roasted and ground coffee sample to show compounds linked to the quality of the beverage.
In the future, consumers will use technology in the supermarket
The artificial intelligence built into the system learns to “look” at the sample, extract patterns and predict the global quality class of the coffee being analyzed. AI is a research area that brings together several disciplines and seeks to develop computational methods capable of imitating human cognitive skills, such as learning, logic and process optimization.
For now, the smart tool is a prototype, but it is already born with high potential to provoke innovation in the coffee value chain. It is believed that, in the future, the system may be used by producers in the field and even by consumers in supermarkets.
Smart computing to understand coffee
Research with CoffeeClass began in 2013 and was developed by Embrapa analyst Ednaldo José Ferreira. The work was carried out within the scope of the Research Café Consortium, created in 1997 and which involves about 50 teaching, research and rural extension institutions, under the coordination of Embrapa Café. Ferreira says that the research arose from a lack of the chain that demanded an automated tool for the analysis of coffee.
The proposal was to generate an automated tool to assist the work of coffee tasters – specialists in sensory analysis – and thus consolidate a technology specifically aimed at certification based on the Global Quality (QG) of the ABIC Coffee Quality Program (PQC). . The proof of concept and the prototype were initially based on the classification of this program, but, according to Ferreira, the technology can be adapted to other indicators of coffee quality.
The HQ is the joint perception of characteristics, such as aroma, flavor, body, acidity, astringency, bitterness, fragrance and others that contribute to the composition of an overall note within a sensory scale - 0 to 10 for the ABIC PQC - used as an indicator of product quality. It is performed by a team of trained specialists who use the standardized preparation of a coffee drink as the basis of the analysis protocol, a process popularly known as "cup proof".
The PQC, launched in 2004, uses this QG score to certify the product in three categories: “Gourmet”, for very high quality coffees; “Superior”, for relatively high quality coffees and “Traditional/Extra strong”, for coffees of recommendable quality and affordable price.
An easy-to-use technology
Doctor in Computer Science and Computational Mathematics, Ferreira explains that several tools were tested before arriving at CoffeeClass. “The focus of the research has always been to generate a simple, low-cost apparatus that could be operated without difficulties, with minimal commands, by all actors in the coffee chain, on the farm itself, before leaving the field, in laboratories, cooperatives, , roasters, supermarkets, coffee shops, among others”, he says.
Ferreira used Data Science to carry out the pilot study with 120 samples of roasted and ground coffee, from various regions of the country, already certified by the PQC, which made it possible to generate a database with more than 2,400 digital images. “We identified the potential and designed a module called the smart core, based on a set of Artificial Intelligence methods and techniques that aim to automatically improve the performance of the analysis”, he explains.
In laboratory tests, the technology identified the Gourmet, Superior, Traditional and Non-Recommended coffee categories with a percentage slightly higher than 75% hit for blind samples.
The study involved low-cost commercial digital microscopes that use different LEDs (light emitting diodes) as a source of illumination for a sample of approximately one gram, in addition to a sample holder and software developed by Embrapa Instrumentação. .
Ferreira explains that the interaction of different lights with the coffee sample produces patterns in images enlarged by digital microscopes, which make it possible to characterize the coffee. The lights at different wavelengths allow detecting the composition of the coffee, as well as mixtures with low quality beans (defects and others), which interfere with the overall quality.
“The color variations observed in enlarged digital images, acquired directly from a sample of roasted and ground coffee, reveal characteristics closely linked to the overall quality of the product”, he explains.
The tool showed significant success potential in laboratory analyses, however, the project leader believes that it is possible to go further and develop other indicators to improve CoffeeClass.
“The challenge now is to understand the characteristics revealed by the intelligent system and expand the horizon of application of the technology”, says Ferreira. To carry out this new stage of the research, he says that he prepared and submitted a new project to the Consórcio Pesquisa Café program, with which he intends to evolve the study.
In the next phase, the work will bring together diverse expertise from universities, research institutes, government agencies, such as the Ministry of Agriculture, Livestock and Supply (Mapa) and its official laboratories (Lanagros), in addition to other historical and recognized entities. excellence in the coffee chain.
“In order to automate and modernize these analyses, Embrapa Instrumentação, with the support of Consórcio Pesquisa Café and the partnership with ABIC, will finalize and launch this technology, unprecedented in the world. The CoffeeClass will undoubtedly revolutionize the quality classification of Brazilian Coffees, so that, more and more, we can consolidate and conquer new markets”, says the head of Technology Transfer at Embrapa Café, Lucas Tadeu Ferreira.
EN 
PT 
