Startup develops computer system for plant sorting and selection

Software installed in electromechanical device sorts cuttings according to growth potential

By José Tadeu Arantes 

Agência FAPESP – A computer system capable of analyzing images of young ornamental plants and classifying them according to their quality is now available in Brazil.

The system was originally designed to sort African violet cuttings but can be adapted for use in classifying other plant species. It was developed by a company called MVisia with support from FAPESP through the Program for Innovative Research in Small Businesses (PIPE). 

“Our vision system scans photographs of the plants and identifies a set of parameters, sorting them with a success rate of up to 80%,” said engineer Luiz Lamardo Silva, who founded MVisia, which is currently being incubated at the Center for Innovation, Entrepreneurship & Technology (CIETEC) attached to the University of São Paulo’s Energy & Nuclear Research Institute (IPEN-USP). 

According to Silva, the high success rate shows that MVisia’s system outperforms basic, intuitive human sorting capabilities by a significant margin. “We developed the system for a midmarket firm in Holambra, where seedlings and cuttings used to be sorted manually by staff. Manual sorting is the rule in this industry. It’s a highly repetitive activity, and accuracy declines sharply after one or two hours of work, giving rise to quite a high error rate,” Silva said. A town in the interior of São Paulo State not far from Campinas, Holambra is the hub of Brazil’s flower and ornamental plant industry.

Because of these errors, plants would frequently be sorted into lots that did not correspond to their true growth potential. Seedlings and cuttings are moved through a series of greenhouses and submitted to different procedures throughout their growth cycle, so wrongly sorted plants would not receive appropriate treatment, and the result was loss of quality in the end product. 

“Human error also led to inefficient use of space,” Silva said. “If a particular plant doesn’t reach the condition required to occupy its place in a greenhouse, the area reserved for it remains empty. The cost of this wasted space is far from negligible, given the need for irrigation, heat regulation and so on.” 

Sorting criteria

Inaccurate sorting criteria, repetitive work and the discomfort of performing the task in a greenhouse led to high rates of labor turnover for MVisia’s customer, hence the need for a system that could standardize and optimize the sorting process.

“The customer was highly satisfied with the results we achieved. The 80% success rate was a major improvement in quality for its operation,” Silva said.

To achieve this level of quality, MVisia’s engineers took as their benchmark a group of 300 cuttings sorted by an expert into four subsets of 75 units each, labeled A, B, C and D. 

They photographed each cutting twice and used computational techniques to extract 26 parameters from the resulting 600 images. They then reduced the number of features to 11 by filtering to remove irrelevant or redundant attributes. The final software was built using these features. 

“The equipment consists of a conveyor belt on which the plants are placed manually. They’re photographed as they pass under a camera, and the images are decoded by the software,” Silva said. 

“Four air nozzles are installed at intervals along the conveyor, one for each of the four classes, A, B, C and D. When a plant passes the nozzle corresponding to its classification, the system switches on a blast of air and pushes the plant off the conveyor into the correct receptacle, which is manually taken to the appropriate greenhouse.” 

The research project proved the feasibility of sorting plants by means of computer vision and artificial intelligence techniques. By comparing different techniques, the engineers were able to pinpoint and combine those best suited to the task. They are now working on an adaptation of the system to sort and select eucalyptus seedlings. 

Brazil is the world’s largest grower of eucalyptus trees. In 2014, it produced almost 16.5 million metric tons of pulp, according to IBA (Indústria Brasileira de Árvores), a tree plantation industry association representing 60 companies and nine state entities. 

“This is an economically powerful industry,” Silva said. “We have the solution it needs. The software is ready to go once a few mechanical issues related to the operational part have been resolved.”