Extraordinary advances over the 10-15 years in precision agriculture and UAS-based remote sensing have proven to be effective tools for collecting and analyzing spatial data for crop growth analytics. In the realm of crop breeding plots, this approach is commonly referred to as "High Throughput Phenotyping." it utilizes sensor systems such as thermal infrared (TIR) and color infrared (CIR) imaging using multi-spectral (MS) sensors & computational tools to extract phenotypic data for large populations. They are used to draw inferences on crop growth parameters, especially when conducting breeding experiments for disease and drought-resistant varieties. However, the major limitation of the widespread adoption of this technology in agriculture is the complex and time-consuming data processing systems and data analysis to estimate numerical targets. The major disadvantage is locating plots precisely in high-resolution imagery of the field and delineating plot boundaries within the Ortho mosaic images of field experiments. Therefore, the goal is to develop a data analytic technique to map row by row RTK GPS quality planting map to rapidly extract plot level soybean canopy reflectance and emittance response to compare and contrast its advantages with manual techniques.

1. Record in-season real-time kinematic (RTK) GPS planting and canopy thermal and multi-spectral (MS) profiles using thermal infrared and multispectral imaging systems.

2. Develop a framework to automatically delineate plots by rows using RTK GPS planting data and extract descriptive data from a Multi-Spectral remotely sensed data layer.

3. Develop data processing technique to map row-by-row RTK GPS quality pertinent features and other relevant metrics map.

4. Evaluate the effectiveness of rapid data extraction and analytic tools using RTK GPS planting data and remotely sensed layer for high-throughput phenotyping and precision agriculture applications.

The proposed project will develop a functional semi-automated pipeline for rapid data analysis of soybean breeding plots, which will involve RTK-GPS surveyed information on test plots, high spatial and RTK-GPS accuracy planted data, novel techniques to utilize as-planted data to develop row-by-row planted maps. The methodology will provide the ability for rapid extraction of plot-by-plot canopy response to screen and analyze hundreds of seed varieties in a large field. The newer data analytic technique is expected to provide efficient, adaptable, and replicable data analysis that minimizes time, labor, and user involvement. The results drawn from the method will be equivalent to or better than manual analysis. This data analytic technique can provide critical knowledge for high throughput phenotyping. It can be utilized in breeding and precision agriculture programs to develop more resilient soybean programs for Kansas growers.

Updated February 20, 2024:
Soybean planting data from a 4-row planter was collected during the 2022 season in soybean breeding trial plots. The planter was equipped with a custom data acquisition system to collect high-frequency RTK GPS, speed, and other machine parameters. UAV imagery was acquired by the five-band multispectral sensing system flown on canopy closure at 30 meters above the ground. Individual camera captures were imported into Agi soft's Meta Shape photogrammetry software to generate orthoimages of the field. Additional data was gathered, including experimental design, field map layout, coordinates of specific areas of interest, and phenotype data. The planter GPS data was utilized to develop a methodology for drawing plot boundaries that are derived without relying on image features and can be drawn regardless of vegetation presence. The plot boundaries with unique labels were established to facilitate row-by-row spectral data extraction for each plot, which will subsequently be employed for downstream rapid analytics.

Updated February 20, 2024:
The imagery was imported in python using raster IO library and its meta data was explored to retrieve information about the raster, understanding the data's properties, and conducting geospatial analysis, or visualizing the raster. Afterwards the boundary images were superimposed on raster to extract row by row spectral signatures of plots along with its unique plot ids, areas, and coordinates. Since the distance between GPS points varied between different plots, the resulting line string lengths differed. To standardize the buffer lengths for each plot, all buffers were resized to the same length. In the last step zonal statistics, table was calculated consisting of plot ID, row ID, unique ID, a polygon object with coordinates of the four plot corners, centroid, area, length, width, pixel count, and VI (Vegetation Index) values. It was observed that the workflow reduced the need for many inputs to adjust plot boundaries. Overlapping canopies or crop lodging from adjacent plots or crop effects do not limit the effectiveness of the workflow. Reproducible workflow was obtained both single-row and multiple-row plot boundaries spectral data collection. The plot boundary extraction methodology presented in the study provides accurate and efficient plot extraction method. This research methodology used simpler existing algorithms to extract spatial signatures from imagery and plot boundary extraction from high accuracy precision planter. In the coming year, we plan to merge field-obtained phenotype plot data with spectral and spatial data files obtained using the above-mentioned pipeline.

View uploaded report

We used data from the farm equipment that plants crops to create a detailed map of where each row of crops was planted. This data, collected at a speedy rate, gave us highly accurate location information thanks to advanced GPS technology. We then turned these planting locations into lines to represent the paths the planting machine took, and we created areas around these lines. Next, we overlaid this planting information onto high-resolution images of the field taken from a low-flying aircraft to see how the crops responded.
We have completed all the necessary steps, from the initial planting data to creating these areas around each row of crops, then overlaid them on pre-cleaned UAV image data to extract plot values. Afterward, we saved them in a special file format. Additional data was gathered, including experimental design, field map layout, coordinates of specific areas of interest, and phenotype data. After all the processing, a file was obtained, which included information about each planting area, like its ID, location, shape, size, and some values that tell us how healthy the plants are. We noticed that this way of doing things meant we didn't need to make many adjustments to the planting areas.
The method we used to figure out where the planting areas are is pretty accurate and doesn't require a lot of complicated steps. We used simple computer tricks to get information from pictures and data from the planting machine to find the planting areas. This automated data pipeline can help farmers and breeders to analyze large-scale plot-level data to extract valuable knowledge on the performance of different breeding varieties. This also helps to analyze greater spatial and temporal resolution data because the time to analyze data has been reduced to minutes from days. Overall, this automated data analytic pipeline can help hasten the process of selecting drought and disease-resistant varieties, which will help growers' operations become more profitable and sustainable in the long term.

Quickly analyzing soybean breeding data, especially for drought and disease-resistant varieties, can accelerate the development of soybean varieties that farmers can adopt promptly. Given unpredictable weather and the impact of diseases like SDS on crop yields, these resistant varieties could boost soybean producers' productivity and profits. Furthermore, remote sensing data is becoming increasingly important for researchers and producers seeking to collect and utilize data. These swift analytical methods can facilitate the easy extraction and analysis of these datasets, enabling valuable insights.