- A team at CRAG develops a novel automatized pipeline for comprehensive phenomic analysis of fruit morphology traits to increase plant breeding efficiency in a fast and economical way.
- The research addresses one of the biggest challenges and opportunities towards increasing the pace of artificial selection and attaining precision agriculture.
- The developed machine learning algorithm can easily be applied to other fruits like apples, tomatoes and citrus, among others.
As worldwide population continues to grow exponentially and climate change increases drought areas, food production needs to be boosted and optimized. Over the last decades, the increase in fruit food systems’ efficiency –one of the main goals of the International Year of Fruits and Vegetables (IYFV) designated by the UN General Assembly– has been achieved thanks to plant breeding programmes that have benefited from the development of genomic technologies. Nevertheless, plant breeding involves both genomics and phenomics (how the genome in different environments leading to differentiated measurable traits), and automatizing phenomic measurements is one of the biggest challenges and opportunities towards increasing the pace of artificial selection and attaining precision agriculture.
Since fruit appearance critically influences consumer acceptance, with different preferences around the world and between communities, morphological traits such as shape, size and colour are highly relevant in many plant breeding programmes. Characterizing such traits manually is costly and inaccurate, but nowadays hundreds of fruit pictures grown under different environmental conditions can be inexpensively taken, even in the field, to collect objective phenomic information. Therefore, developing new and improved analytical tools capable of automatically transforming this wealth of imaging data into valuable knowledge is key to enhance fruit appearance evaluation.
CRAG researchers have developed an automatized and cost-effective computing method to evaluate fruit shape and colour that will contribute to increase agriculture efficiency. The study, published at the scientific journal Plant Phenomics, has been carried out using strawberry images, although its machine learning algorithm can easily be applied to other fruits like apples, tomatoes and citrus. The devised software pipeline is also able to predict fruit shapes and appearance, providing a powerful simulating tool to design new crosses. The researchers have given open access to the code for the community to adapt it to their own needs.
Implementing deep learning algorithms
In this study, researchers took external and half-cut pictures of about 2000 strawberry fruits from different breeding lines provided by the Planasa company, harvested in the 2018 campaign in Huelva (Spain), the main European strawberry producing area. “Evaluating the shape of a given object, a strawberry in this case, from its picture is not as straightforward as it may seem. Classical linear descriptors –area, perimeter, height, width…– have certain limitations, leading to the loss of relevant information by extremely simplifying morphology features. To better assess shape, we complemented these linear methods with multivariate and deep learning techniques”, explains the first author of the article, Laura M. Zingaretti, who has carried out this work as part of her doctoral thesis at CRAG.
For the first time, this work applies deep learning techniques, a class of machine learning algorithms, to evaluate fruit shape. Combining such methods with lineal and multivariate measurements, researchers were able to generate an automatized software pipeline that analyses shape and colour patterns extracted from strawberry images. The developed tool is quite more automatized than their predecessors as it requires minimal user intervention and limited computer time, providing an inexpensive and fast way for phenomic evaluation.
A tool for improving agricultural efficiency
“In addition to the morphological analysis, our deep learning tool presents a novel idea to simulate new fruit shapes on the computer, since it is capable of predicting the appearance of the fruits of new crosses. This contribution can be very valuable in the first step of breeding programmes, since it would allow evaluating various crosses without the need to test them directly in the field, saving time and resources”, points out Miguel Pérez-Enciso, ICREA researcher at CRAG co-directing the thesis.
Overall, the developed pipeline shows that fruit shape and colour can be quickly and automatically evaluated and are quite heritable, which will allow breeders to rapidly make decisions to modify conformational traits of agricultural products. “This study has a direct impact on the agricultural sector since the algorithms are designed to obtain morphological parameters in an efficient and economical way. Additionally, this tool has the potential to be adapted to measure visual fruit phenomic traits directly in the field, to analyse other plant conformation characteristics (leaves, flowers, roots…), or for early disease assessment”, adds Amparo Monfort, IRTA researcher at CRAG and co-director of the work.
Image credits:
Pictures kindly provided by CRAG, and re-used with permission.