Overview / Usage

The common bean (Phaseolus vulgaris L.) is one of the most important species within the group of edible legumes in America and Africa. (Chacón, Pickersgill, & Debouck, 2005). Along with corn and cassava, it is one of the oldest cultivated species in the new world (Broughton et al., 2003). Actually, it has a world production of over 26 million metric tons per year, with Asia and Latin America being the main producers (FAO, 2016). It is consumed mainly in cooked grain; however, it is also consumed in a large percentage as green bean and pod (Gepts, 2001). It is an important source of protein (protein content in seeds between 20 and 25%) and minerals, which represents a significant contribution to human nutrition (Broughton et al., 2003).
The climate change is affecting the normal development of the crops and is causing irreversible damage into the metabolism of plants, this is the mainly reason why, in the last few years researchers and plant breeders have focused on developing smart crops varieties that are able to adapt to the current conditions. In beans particularly, a common consequence of this phenomenon, is the heat stress, this abiotic factor affects directly the pollen viability, additionally cause large losses in yield. Tolerant genotypes and heat-sensitive genotypes of Phaseolus vulgaris contrast strongly with respect to pollen morphology, pollen viability and pollen wall architecture when subjected to high temperatures during microsporogenesis, so this characteristic should be evaluated in order to identify resistant genotypes to the high temperatures.
The current pollen viability evaluation techniques are very time consuming, since the expert needs to count manually and classify each grain of pollen, while his is observing the sample in the microscope, could you imagine doing the same task 500 times?

Methodology / Approach

In collaboration with staff from the Andean Bean Program at CIAT, we created software with OpenCV and Accord.Net to automatically evaluate the pollen viability using images taken in the microscope. This software was trained with around 1000 images from viable and non-viable pollen grains, and extracted features such as width, height, color, and roundness. Different segmentation algorithms and machine learning models were evaluated.

Other links

• Photo Measure ( Pollen Viability)