Founder & CEO Cropper Agricultural Technologies Inc.
Disclosure: Disclosure information not submitted.
High-throughput phenotyping for plant breeding plays a vital role in creating new plant varieties to feed our ever-growing population and to improve the agricultural economy. Hence, for monitoring plant growth, health, and traits, automated analysis of available aerial images is an attractive alternative to time-consuming manual inspection. A fundamental step of these phenotypic studies on field trials is to detect and localize research microplots within the field. A naive approach to solving this problem is to generate bounding boxes based on the known geometry of the field. However, bounding boxes that are generated in this way are often misaligned from the actual position of the microplots on orthomosaic images of field trials due to inaccuracies in geo-location information on the drone and errors during the image stitching process when generating the orthomosaic. To overcome this, we have developed a two-step optimization method. First, the initial bounding boxes area is optimized using an objective function that maximizes the level of vegetation inside the area. We then adjusted the position of microplots on an individual basis using an objective function that simultaneously maximizes the area of the microplot overlapping vegetation, minimizes spacing variance between microplots, and minimizes each plot’s translation relative to other microplots in the same row and column. The orthomosaics used in this study were obtained from multiple dates of a canola and wheat breeding trials during 2017-2018. We compared the positions of automatically refined plot segmentations to ground truth segmentations using Dice similarity coefficient (DSC), and achieved an average DSC of 92% and 82.57% across all microplots and orthomosaics tested for canola and wheat datasets, respectively. Furthermore, the algorithm was able to accurately extract 100% of microplots for canola and 96.26% for wheat.