Blog: Discussing AI applications in Agriculture in the weekly team meeting
Last week on Wednesday, October 2, I (Divya) presented a short overview presentation on the application of AI (or technology, in general) to Agriculture in the weekly team meeting of I2SC. Coming from Nepal, primarily an agricultural country, I was interested in learning how technology could be applied to Agriculture. I based my presentation on the works in agriculture mentioned in the 2020 survey paper titled "Artificial Intelligence for Social Good". Even though a bit old, I would still recommend the survey paper to get an overview of AI applications in different areas of Society. By 2080, we would've ~2B more people to feed [1] but already in 2023, 2.33 billion people struggled to access adequate food regularly and about 864 million people faced severe food insecurity [2]. Moreover, about 28% of the employed population of the world depends on Agriculture [3], thus we must look into the techniques to improve food production.
The survey paper mentions the works in categories such as crop planning, crop maintenance, mitigating crop disease, crop yield prediction, and agricultural information gathering. For crop planning, the paper mentions the work by Bruneli and Luecken (2009) [4] where they use multi-objective evolutionary algorithms to determine the optimal crop using objectives like minimizing the cost of fertilizing, liming, cultivation, and maximizing the total returns for the farmers along with the sustainable use of the land. For crop maintenance, the paper mentions how Evapotranspiration (ET) tracking devices can be used in building efficient irrigation systems. We also learned how a Lysimeter, which is a device used to measure Evapotranspiration, works. Holman et. al. (2013) [5] use Gaussian-process models and Neural Networks to relationship between non-ET weather station data and reference ET from ET networks. In mitigating crop disease, we checked the work by Quinn et. al. (2011) [6] in determining optimal routes for surveying the plants affected by Cassava disease. Then, we checked the work by You et. al. (2017) [7] who use a novel dimensionality reduction technique to train CNN/LSTM models in low-resource settings. They also use a Gaussian process model and show that their approach outperformed previous methods in county-level soybean yield prediction. Finally, the paper showed how Swamy et. al. (2019) [8] created a low-cost aerial device using helium balloons and a phone to gather agricultural information for smallholder farmers. Overall, I found the experience of learning about such AI applications and presenting to the team, quite fun.
[1] https://population.un.org/wpp/
[2] https://www.fao.org/interactive/state-of-food-security-nutrition/en/
[3] https://www.ilo.org/industries-and-sectors/agriculture-plantations-other-rural-sectors
[4] https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/2212
[5] https://www.ijcai.org/Proceedings/13/Papers/415.pdf
[6] https://cdn.aaai.org/ojs/7811/7811-13-11339-1-2-20201228.pdf
[7] https://cs.stanford.edu/~ermon/papers/cropyield_AAAI17.pdf
[8] https://dl.acm.org/doi/10.1145/3314344.3332485
