The detection of wheat heads in plant images is an important task for estimating pertinent wheat traits including head population density and head characteristics such as health, size, maturity stage, and the presence of awns. Several studies have developed methods for wheat head detection from high-resolution RGB imagery based on machine learning algorithms. However, these methods have generally been calibrated and validated on limited datasets. High variability in observational conditions, genotypic differences, development stages, and head orientation makes wheat head detection a challenge for computer vision. Further, possible blurring due to motion or wind and overlap between heads for dense populations make this task even more complex. Through a joint international collaborative effort, we have built a large, diverse, and well-labelled dataset of wheat images, called the Global Wheat Head Detection (GWHD) dataset. It contains 4700 high-resolution RGB images and 190000 labelled wheat heads collected from several countries around the world at different growth stages with a wide range of genotypes. Guidelines for image acquisition, associating minimum metadata to respect FAIR principles, and consistent head labelling methods are proposed when developing new head detection datasets. The GWHD dataset is publicly available at http://www.global-wheat.com/and aimed at developing and benchmarking methods for wheat head detection.

Plant breeding trials are extensive (100s to 1000s of plots) and are difficult and expensive to monitor by conventional means, especially where measurements are time-sensitive. For example, in a land-based measure of canopy temperature (hand-held infrared thermometer at two to 10 plots per minute), the atmospheric conditions may change greatly during the time of measurement. Such sensors measure small spot samples (2 to 50 cm²), whereas image-based methods allow the sampling of entire plots (2 to 30 m²). A higher aerial position allows the rapid measurement of large numbers of plots if the altitude is low (10 to 40 m) and the flight control is sufficiently precise to collect high-resolution images. This paper outlines the implementation of a customized robotic helicopter (gas-powered, 1.78-m rotor diameter) with autonomous flight control and software to plan flights over experiments that were 0.5 to 3 ha in area and, then, to extract, straighten and characterize multiple experimental field plots from images taken by three cameras. With a capacity to carry 1.5 kg for 30 min or 1.1 kg for 60 min, the system successfully completed >150 flights for a total duration of 40 h. Example applications presented here are estimations of the variation in: ground cover in sorghum (early season); canopy temperature in sugarcane (mid-season); and three-dimensional measures of crop lodging in wheat (late season). Together with this hardware platform, improved software to automate the production of ortho-mosaics and digital elevation models and to extract plot data would further benefit the development of high-throughput field-based phenotyping systems.

Climatic variability in dryland production environments (E) generates variable yield and crop production risks. Optimal combinations of genotype (G) and management (M) depend strongly on E and thus vary among sites and seasons. Traditional crop improvement seeks broadly adapted genotypes to give best average performance under a standard management regime across the entire production region, with some subsequent manipulation of management regionally in response to average local environmental conditions. This process does not search the full spectrum of potential G × M × E combinations forming the adaptation landscape. Here we examine the potential value (relative to the conventional, broad adaptation approach) of exploiting specific adaptation arising from G × M × E. We present an in-silico analysis for sorghum production in Australia using the APSIM sorghum model. Crop design (G × M) is optimised for subsets of locations within the production region (specific adaptation) and is compared with the optimum G across all environments with locally modified M (broad adaptation). We find that geographic subregions that have frequencies of major environment types substantially different from that for the entire production region show greatest advantage for specific adaptation. Although the specific adaptation approach confers yield and production risk advantages at industry scale, even greater benefits should be achievable with better predictors of environment-type likelihood than that conferred by location alone.