A Systematic Review of the Literature on Machine Learning Methods Applied to High Throughput Phenotyping in Agricultural Production

The amount of images that can be extracted from crops such as soybean, corn, sorghum, etc., has increased exponentially due to the proliferation of remote sensing technologies such as Unmanned Aerial Vehicles (UAV). When processed and analyzed, images can provide valuable information and knowledge about High Throughput Phenotyping (HTP). Advances in HTP technology are essential to ensure that crop genetic improvement meets future global demands for food and fuel. In addition to UAVs, Digital Image Processing (DIP) and Machine Learning (ML) methods have shown to be promising tools in HTP to minimize the time and cost of analyzing entire crops. However, the performance and quality of the results obtained in HTP depend on the techniques used throughout the process. With this limitation in mind, the objective of this article is to present a Systematic Literature Review (SLR) on image capture techniques, DIP and ML applied to HTP. This review focuses on four sources of scientific searches, which initially returned 161 articles to be analyzed, of which 46 were excluded due to the Exclusion Criteria (EC), and 43 were duplicates, leaving only 72 for full reading. Of the 72 articles read, 27 were excluded due to the Exclusion and Quality Criteria (QC). Finally, 45 studies remained, resulting in a useful base on the cameras/sensors used in capturing the images, the most analyzed agronomic traits in the crops, in addition to a survey on the main DIP, ML techniques used in HTP.