How AI Revolutionizes Established Tech — Part 3
One would not immediately associate Moore’s Law with crop production, nevertheless having annual yield improvements mimic the astonishing progress the semiconductor industry enjoyed in the late-twentieth and early twenty-first centuries is an appealing goal for crop scientists. To attain similar ambitious targets for greenhouses and in general plant production, crop researchers are turning to AI algorithms that go beyond applications to genomic sequencing. In Part 3 of our Series: How AI Revolutionizes Established Tech, we have a look at how AI can help crop scientists understand how plants follow genomic instructions. It is not enough for a plant to exhibit potential, one wants to know how that potential comes to be, so that it can be capitalized on.
Part 3: AI Watches Plants Grow
Plants for commercial use or consumption have been traditionally grown by starting with a specific breed and then adhering to strict growing protocols. The result was a mixed crop of plants of different leaf size, skin-cell density and metabolic ability. The changes observed in the plants’ physical traits of phenome were mostly due to varying light levels and plant handling.
To surpass that, crop scientists wanted to be able to align phenotypic evaluations of plants with their genetic potentials. That meant observing plants grow, thereby leaving behind established techniques such as optical imaging. New techniques had to emerge where researchers would develop software algorithms for smartphone cameras that would allow growers to quantify parts of the plant’s phenotype. These algorithms would work hand-in-hand with data collected from hyperspectral, fluorescent and tomographic sensors fitted on drones. Images of test crops would be annotated, and labels added for identification of specific plant characteristics. An AI algorithm would then be trained on the traits of a specific plant such as, for example, flower bud count on orchids.
This scientific discipline is now known as crop phenomics research. It combines agronomy, life sciences, information science, math and engineering. Additionally, it makes use of high-performance computing and artificial intelligence technology to explore diverse phenotypic information of crop growth in complex environments. The essential outcome is to construct an operational system able to phenotype crops in a high-throughput, multi-dimensional and automatically measuring manner, and create a tool that systematically integrates big data from a multi-modality and multi-scale genotypic condition.
Using AI instead of traditional techniques where plants are described manually saves time, as an image analysis algorithm can detect for example a bud in a greenhouse full of plants with increased speed and accuracy. What was once a challenge given the need for high resolution imaging of samples in view of the output on a micron level has now been replaced with section imaging where phenotypic information is being extracted in sequence, and then processed by AI software.
These days, AI techniques allow the development of new methods of mining genes of interest such as those associated with important agronomic traits. Due to these improved analysis algorithms, new intelligent solutions can be proposed for precision breeding. Still, phenotyping at tissue and cellular scale continues to require complex procedures, which calls for the need to simplify the sample preparation process and a continued exploration of advanced imaging and AI techniques, in order to accelerate the microscopic phenotyping procedures.