It is essential to adopt sustainable agricultural intensification practices in existing croplands to meet this demand, which is expected to rise significantly by 2050 as a result of population growth, rising per capita income, and the expanding use of biofuels.
The estimating methods used today in the global South are still insufficient. Self-reporting and crop cutting are two conventional procedures that have limits, and remote sensing technologies are not completely employed in this situation.
However, new developments in machine learning and artificial intelligence, notably deep learning with convolutional neural networks (CNNs), provide hopeful solutions in this regard. Researchers from Japan carried out a rice-focused investigation to determine the extent of this new technology.
To predict rice yield, they integrated ground-based digital photographs acquired at the crop’s harvesting stage with CNNs. Their research was published in Volume 5 of Plant Phenomics on July 28, 2023, and it was made available online on June 29, 2023.
“We got things going by running a significant field campaign. In order to build a comprehensive international database, we collected rice canopy images and rough grain yield data from 20 locations across seven different countries, according to Dr. Yu Tanaka, associate professor at the Graduate School of Environmental, Life, Natural Science and Technology at Okayama University and the study’s principal investigator.
Digital cameras were used to take pictures because they could collect the necessary information at a distance of 0.8 to 0.9 meters, vertically downward from the rice canopy.
The team successfully developed a database of 4,820 yield data of harvesting plots and 22,067 images, encompassing various rice cultivars, production systems, and crop management techniques, with Dr. Kazuki Saito of the International Rice Research Institute (formerly Africa Rice Center) and other collaborators.
After that, a CNN model was created to calculate the grain yield for each image that had been gathered. To see how different sections in the photos of the rice canopy added up, the team utilized a visual-occlusion technique.
It required masking particular areas of the photos and tracking how the model’s yield estimation changed as a result. Understanding how the CNN model interpreted different aspects in the photos of the rice canopy helped the researchers improve their capacity to discern between parts of the canopy that contribute to yield and others that do not.
In the validation and test datasets, the model performed well, explaining 68%–69% of the variation in yield. The findings of the study demonstrated the significance of panicles, loosely branched clusters of flowers, in estimating yield using occlusion-based visualization.
Must read: AI is Impacting Candidate Screening and Sourcing
The model was able to identify mature panicles in the ripening stage and forecast production properly. It was also able to identify cultivar and water management differences in yield in the prediction dataset. But when image resolution dropped, it became less accurate.
The model nevertheless shown strength, displaying good accuracy at various shooting angles and times of day. Overall, the created CNN model showed promise in estimating rough grain yield from photos of the canopy of rice in various conditions and genotypes. In addition to not requiring labor-intensive crop trims or sophisticated remote sensing technology, it is also very cost-effective, adds Dr. Tanaka.
The work highlights the possibility of using CNN-based models to track rice productivity at a regional level. Further study should concentrate on modifying the model to account for poor yielding and rainy areas because the model’s accuracy may vary according on the environment.
The AI-based approach has also been made accessible to farmers and researchers via a straightforward smartphone application, considerably enhancing the technology’s accessibility and its practical applications. ‘HOJO’ is the name of the app, and both iOS and Android already support it.
The researchers anticipate that their study will aid quicker breeding programs and better rice field management, which will benefit global food production and sustainability projects.
Must read: According to a US court, AI-generated art cannot be given copyrights.