One key issues regarding coffee production is the bean selection process, which is a tedious task that nowadays is often manually done. Since traditional selection techniques are based on size and density, most of them do not consider faded, mottled, and blotchy beans. However, beans with these characteristics should be discarded because they affect coffee quality.

The latest technology employed for coffee beans selection uses laser equipment. However, these kinds of machines a very expensive, and therefore they are only used by large producers. Due to the fact that small producers do not have access to this technology, their market competitiveness is drastically reduced.

Current technology favors the use of low cost tools that can be applied to automate and improve the coffee beans selection process. The development of such a tool is the aim of our proposal, which lies on a computer vision system that is capable of identifying the key characteristics (color, shape, size) of coffee beans. Once this task is done, beans are classified according to their quality. Specifically, the low-cost automatic coffee bean classifier has the following main stages:

Sieving (A): First, coffee beans are classified by size using a traditional sieving method, which is important because it prevents blocking of the proposed mechanism in the funneling stage.

Funneling (B): The objective of this stage is to organize the beans in a row. To this end, coffee beans of similar size are transported to a hopper that is connected with a vibratory funnel with appropriate diameter (the vibration mechanism prevents beans from getting stuck in the funnel). By gravity, the beans arranged in a row are placed on a conveyor belt. This step is necessary to facilitate the visualization and selection process.

Visualization (C): Once the beans are being transported on the conveyor belt, an embedded system equipped with a camera identifies key characteristic (related to color and shape) of each coffee bean. Based on machine-learning, the embedded system is capable to decide the beans quality level.

Selection (D): The previously described embedded system sends a control signal to a mobile stick that directs the beans towards two different hoppers according to the coffee quality.

The components of the selection mechanism can be effortlessly acquired at low cost since they have been broadly employed in different industrial and electronics applications. Indeed, we can use the last generation of low-cost electronic programmable devices for around $30 (Arduino and Raspberry Pi devices). In summary, acquirement, implementation, and assemblage of our proposed components are not problematic.

Although our proposal is focused on coffee beans, it can be also suitable for quality selection processes of other grains such as seeds and legumes. In addition to agricultural applications, industrial and military applications could be a new target of our proposed mechanism.

In summary, our proposal consists in a low-cost device for quality selection processes, which is accessible for small producers.