Technique using light and artificial intelligence is effective in selecting immature soybean seeds

Research conducted by Brazilian scientists has demonstrated that chlorophyll fluorescence is a reliable predictor of seed maturity. The greener the seed, the lower the quality. Manual separation is the rule in Brazil.

By Luciana Constantino  |  Agência FAPESP – Historically based on tradition and experience, the decision-making process in agriculture has been transformed in recent years by technological innovations that scale up production and provide solutions to the challenges posed by pests, natural limitations on arable land and the effects of climate change.

Brazilian researchers have developed a technique to help select seeds of soybeans and other legumes in accordance with maturity stages, assuring physiological quality without destroying samples.

The scientists used light and artificial intelligence (AI) to show that chlorophyll fluorescence is an effective and reliable indicator of soybean seed maturity. They validated the results by means of machine learning algorithms. The novel technique can be used to classify commercial seeds.

The greener and less mature the seeds, the less vigor and germinating power they have, so that their quality is lower. As a result, the market value of soybean seed lots with more than 8% green seeds is reduced and they cannot be exported. Green seeds also produce less oil, with higher acidity and higher refining costs.

Manual seed quality analysis is required by law in Brazil. It must be performed by a technician accredited with the Ministry of Agriculture and entails visual separation based on color. Green seeds are discarded and destroyed, forming waste. 

“I consider this study a milestone. No studies in the literature to date have addressed the possibility of separating seed stages based on chlorophyll fluorescence. The method can be used for other legumes besides soybeans. It’s a major advance in scientific knowledge,” said Thiago Barbosa Batista, first author of an article on the study published in the journal Frontiers in Plant Science.

The research was part of Batista’s PhD thesis, developed with FAPESP’s support. His thesis advisor was Edvaldo Aparecido Amaral da Silva, a professor at São Paulo State University's School of Agricultural Sciences (FCA-UNESP) in Botucatu, and last author of the article.

“Phenotyping various kinds of seed was the main reason for starting our thematic group. We focused on chlorophyll retention and its association with low quality, and this in turn led to the need to analyze the stages of seed development. The results of this study enhance the reliability of maturity characterization when seeds are similar shades of green, especially in nearby stages,” said Amaral da Silva, who leads a project on the “green seed problem”.

The study was conducted in partnership with Clíssia Barboza da Silva, a researcher at the Radiobiology and Environment Laboratory belonging to the University of São Paulo's Center for Nuclear Energy in Agriculture (CENA-USP). Barboza da Silva is also supported by FAPESP via three projects (17/15220-7, 18/03802-4, and 18/01774-3).

“This technique avoids destroying seeds, which are classified automatically by the AI algorithm. We currently analyze samples, but it could be done seed by seed in future,” she said.

For some years Barboza da Silva has analyzed seeds using light-based technologies such as autofluorescence spectral imaging. In September 2021, a study led by her showed that images based on autofluorescence could be used to detect changes in the optical properties of soybean seed tissue and consistently distinguish between seeds with high and low vigor. An article on the study was published in Scientific Reports.

Maturity in images

The researchers sowed soybean seeds in pots, maintaining relative air humidity at 65% and average air temperature at 24.2 °C. Pods were collected manually during the maturation phase, and the seeds were classified by reproductive stage, as R7.1 (start of maturation), R7.2 (mass maturity), R7.3 (seed disconnected from mother plant), R8 (harvest point), or R9 (final maturity).

Physical parameters, germination, vigor and pigment dynamics were analyzed for seeds collected at different stages of maturation.

High-resolution autofluorescence spectral images (2192x2192 pixels) were captured using a VideometerLab4 system with light-emitting diodes (LEDs) at different excitation wavelengths combined with long-pass optical filters. 

Autofluorescence signals were extracted from images captured at different excitation/emission combinations, but the researchers concluded that the combinations 660/700 nanometers (nm) and 405/600 nm performed fastest and most accurately in identifying the different stages of seed maturation.

Chlorophyll is highly fluorescent. It emits light when exposed to radiation at specific wavelengths because it does not use all the energy from the light and “loses” part of it via fluorescence. This “surplus” is captured by the equipment, which converts it into an electrical signal, generating an image with varying shades of gray as well as white and black. The lighter the area, the higher the chlorophyll content, showing that the seed is less mature.

Mature seeds normally retain chlorophyll as a source of energy while the nutrients required for development of the young plant (lipids, proteins and carbohydrates) are being stored. After fulfilling this function, the chlorophyll degrades, and the less chlorophyll remains, the more advanced the seed is in the maturation process, with more nutrients and better quality.

The “green seed problem” refers to chlorophyll retention in mature seeds and is associated with lower oil and seed quality. It can be caused by frost but is exacerbated by the high temperatures and water stress brought by climate change in recent years.

The article “A reliable method to recognize soybean seed maturation stages based on autofluorescence-spectral Imaging combined with machine learning algorithms” is at: www.frontiersin.org/articles/10.3389/fpls.2022.914287/full.