Challenges
Modern agriculture generates large, complex data sets from unmanned aerial vehicles (UAVs), environmental sensors, ‘omics data, and more. The data collection, processing, and analysis of these data is often complex, labor intensive, and requires careful consideration of the underlying organisms.
Both laboratory and in-the-field data pose unique challenges for data collection, processing, and analytics, including:
- Destructive sampling techniques
- Measurement noise
- Environmental interference
- Labor-intensive data processing
- High-dimensional data sets
- Information-rich, and more complex time series data
Our solutions
We leverage signal processing, shape analytics, data fusion, and artificial intelligence with these rich data to detect pest presence, monitor environmental conditions for efficient resource utilization, and take a deeper look beneath the soil or cell wall. For experimentalists, we provide algorithms for identification of discerning genes in differing physiological conditions, analyzing high-dimensional data from biological experiments and novel biological circuit design. We develop and validate automated data processing algorithms for root structures, and have significant expertise in UAV data analysis for event detection.
Case Study
Automated Root Growth Analysis for Temporal Phenotyping
A significant portion of Earth’s plant biomass exists underground as root systems. Root systems make up an important fraction of plant biomass, and various developmental phenotypes of roots can be interrelated with environment factors that impact both the biome and the individual plant. Roots play critical roles in ecosystems and agriculture; yet, they are occluded and difficult to study.
The quantification of root growth dynamics is an important task in the study of many terrestrial biomes. Minirhizotron systems enable non-destructive imaging of root systems over time and in situ. However, minirhizotrons offer an incomplete picture of an individual plant’s root system, and estimated phenotypes from 2D image slices are merely samples of the true 3D phenotype. Reliable quantification of root system phenotypes is increasingly dependent on data volume as variability of the particular feature in minirhizotron images increases.
The prospect of high-throughput minirhizotron image analysis enabled by true automation is extremely impactful because the necessary sampling volume grows as the rhizotron’s window into the whole plant root system decreases. The impact of possible automation is further amplified by the labor intensive process of analyzing in-situ minirhizotron images in the present state-of-the-art. Automation enables more high-frequency image sampling. With better time resolution temporal registration of root phenotypes becomes easier, in turn leading to even more reliable automation: a virtuous cycle.
We develop and validate algorithms that automatically identify and featurize root structures in minirhizotron images. Leveraging our expertise in computational topology, geometry, and modern development practices, GDA’s automated minirhizotron analysis pipeline enables low cost, high-throughput, in-situ minirhizotron experiments by emphasizing software autonomy and portability, and it removes the need for direct human intervention in root segmentation/classification and analysis. These solutions reduce minirhizotron research costs, enable more efficient use of personnel, and decrease turn-around time on new experiments and data collection efforts.
For large-scale field deployments, GDA has extensive expertise in sensor field coverage determination and relocation tasking for large-scale problems like the minirhizotron sensors that must be distributed widely and strategically to analyze environmental effects. Models to address how complex hydro-geochemical systems respond to adverse conditions will require data from specific locations, and this is our specialty.