Search

Plant population counts are highly valued by crop producers as important early-season indicators of field health. Traditionally, emergence rate estimates have been acquired through manual counting, an approach that is labor-intensive and relies heavily on sampling techniques. By applying deep learning-based object detection models to aerial field imagery, accurate plant population counts can be obtained for much larger areas of a field. Unfortunately, current detection models often perform poorly when they are faced with image conditions that do not closely resemble the data found in their training sets. In this paper, we explore how specific facets of a plant detector’s training set can affect its ability to generalize to unseen image sets. In particular, we examine how a plant detection model’s generalizability is influenced by the size, diversity, and quality of its training data. Our experiments show that the gap between in-distribution and out-of-distribution performance cannot be closed by merely increasing the size of a model’s training set. We also demonstrate the importance of training set diversity in producing generalizable models, and show how different types of annotation noise can elicit different model behaviors in out-of-distribution test sets. We conduct our investigations with a large and diverse dataset of canola field imagery that we assembled over several years. We also present a new web tool, Canola Counter, which is specifically designed for remote-sensed aerial plant detection tasks. We use the Canola Counter tool to prepare our annotated canola seedling dataset and conduct our experiments. Both our dataset and web tool are publicly available.

Increasing crop trait diversity in oilseed rape (OR, Brassica napus L.) cropping systems by introducing frost-sensitive legume species could improve weed suppression and crop productivity. Intercrops and sole crops were compared over two years in the field in Western France. Winter OR was intercropped simultaneously with either spring faba bean (Vicia faba L.) or common vetch (Vicia sativa L.) in a row replacement design without herbicides. Each species was sown at 50% of the recommended sole crop density in alternate rows. Due to the high values of faba bean aboveground traits (height, leaf area, and biomass) and the strong competitive ability for soil N of OR, both species appeared complementary in resource utilization, and thus less soil N and light were available for weeds. The OR–faba bean intercrop was able to reduce weed biomass by 41% compared to the OR–common vetch intercrop. Furthermore, growth and competitive ability of OR for soil N were increased when intercropped with faba bean. Both grain yield and number per plant were three times higher in OR–faba bean intercrops compared to OR sole crops. Under high weed infestation, the presence of faba bean with OR reduced weed aboveground biomass by 35% and weed N accumulation by 11% compared to the OR sole crop. No change was observed in the weed community composition. We observed that a level of aboveground biomass greater than 2 t ha−1 and a soil N uptake at approximately 80 kg ha−1 was needed to reduce biomass and N content of weeds.

Canola (Brassica napus), with its prominent yellow flowers, has unique spectral characteristics and necessitates special spectral indices to quantify the flowers. This study investigated four spectral indices for high-resolution RGB images for segmenting yellow flower pixels. The study compared vegetation indices to digitally quantify canola flower area to develop a seed yield prediction model. A small plot (2.75 m × 6 m) experiment was conducted at Kernen Research Farm, Saskatoon, where canola was grown under six row spacings and eight seeding rates with four replicates (192 plots). The flower canopy reflectance was imaged using a high-resolution (0.15 cm ground sampling distance) 100 MP iXU 1000 RGB sensor mounted on an unpiloted aerial vehicle (UAV). The spectral indices were evaluated for their efficiency in identifying canola flower pixels using linear discriminant analysis (LDA). Digitized flower pixel area was used as a predictor of seed yield to develop four models. Seventy percent of the data were used for model training and 30% for testing. Models were compared using performance metrics: coefficient of determination (R2) and root mean squared error (RMSE). The High-resolution Flowering Index (HrFI), a new flower index proposed in this study, was identified as the most accurate in detecting flower pixels, especially in high-resolution imagery containing within-canopy shadow pixels. There were strong, positive associations between digitized flower area and canola seed yield with the peak flowering timing having a greater R2 (0.82) compared to early flowering (0.72). Cumulative flower pixel area predicted 75% of yield. Our results indicate that the HrFI and Modified Yellowness Index (MYI) were better predictors of canola yield compared to the NDYI and RBNI (Red Blue Normalizing Index) as they were able to discriminate between canola petals and within-canopy shadows. We suggest further studies to evaluate the performance of the HrFI and MYI vegetation indices using medium-resolution UAV and satellite imagery.

Phenotyping crop performance is critical for line selection and variety development in plant breeding. Canola (Brassica napus L.) flowers, the bright yellow flowers, indeterminately increase over a protracted period. Flower production of canola plays an important role in yield determination. Yellowness of canola petals may be a critical reflectance signal and a good predictor of pod number and, therefore, seed yield. However, quantifying flowering based on traditional visual scales is subjective, time-consuming, and labor-consuming. Recent developments in phenotyping technologies using Unmanned Aerial Vehicles (UAVs) make it possible to effectively capture crop information and to predict crop yield via imagery. Our objectives were to investigate the application of vegetation indices in estimating canola flower numbers and to develop a descriptive model of canola seed yield. Fifty-six diverse Brassica genotypes, including 53 B. napus lines, two Brassica carinata lines, and a Brassica juncea variety, were grown near Saskatoon, SK, Canada from 2016 to 2018 and near Melfort and Scott, SK, Canada in 2017. Aerial imagery with geometric and radiometric corrections was collected through the flowering stage using a UAV mounted with a multispectral camera. We found that the normalized difference yellowness index (NDYI) was a useful vegetation index for representing canola yellowness, which is related to canola flowering intensity during the full flowering stage. However, the flowering pixel number estimated by the thresholding method improved the ability of NDYI to detect yellow flowers with coefficient of determination (R2) ranging from 0.54 to 0.95. Moreover, compared with using a single image date, the NDYI-based flowering pixel numbers integrated over time covers more growth information and can be a good predictor of pod number and thus, canola yield with R2 up to 0.42. These results indicate that NDYI-based flowering pixel numbers can perform well in estimating flowering intensity. Integrated flowering intensity extracted from imagery over time can be a potential phenotype associated with canola seed yield.

The field pea has both semi-leafless (SL) and leafed (L) types. Mixing these two types together might improve yield by optimizing pea solar radiation interception, reducing lodging, and decreasing disease. However, an optimum mixing ratio has not yet been established, since previous studies mixed two leaf types from two separate varieties. This study used four near-isogenic pairs of pea genotypes differing only in leaf type to determine the optimal mixing ratio for yield and agronomic traits. Two leaf types were mixed at seeding in five mixing ratios: 0:100, 50:50, 67:33, 83:17, and 100:0 SL/L. With precise UAV quantification of canopy height (r2 = 0.88, RMSE = 2.6 cm), the results showed that a ratio of over 67% semi-leafless pea had a 10% greater lodging resistance when compared to the leafed monoculture. For mycosphaerella blight and Uromyce viciae-fabae rust diseases, the 83:17 mixture decreased disease severity by 4% when compared with the leafed monoculture. Regression analysis of yield estimated that the 86:14 ratio provided an 11% increase to the leafed monoculture, but there was no increase compared with the semi-leafless monoculture. Mixing the two types in a high semi-leafless ratio can reduce leafed lodging and prevent yield loss but does not increase the overall yield over the semi-leafless monoculture.

Traditional methods of estimating frost damage to crops are labor-intensive and time-consuming. Remote sensing imagery and vegetation indices can be used for condition assessment, however, the utility of using vegetative indices in assessing frost damage specifically is not known. The objective of this study was to estimate the freezing injury using a vegetative index developed from hyperspectral imagery. Three replicates of six 6-leaf stage canola plants were subjected to a temperature of −10 °C for 6hr. The resulting frozen plants were imaged at 6 different thawing times using an imaging spectrophotometer (400-1000 nm). Normalized difference vegetation index (NDVI) and triangular vegetation index (TVI) were calculated. Contrary to expectations, NDVI values from frozen plants increased in the initial thawing treatments (1, 2 and 4hr) and then decreased at 8hr. Whereas, TVI values decreased gradually with increased duration of thawing. Furthermore, when compared to NDVI, TVI clearly differentiated frozen from control plants and within the freezing treatments. The differential modes of response of the indices to post-freezing reflectance changes is the possible reason. These results suggest that with recent advancements in low altitude remote sensing in the areas of spectral, spatial and temporal resolution, early estimation of frost damage is possible.

Hailstorms are a frequent natural weather disaster in the Canadian Prairies that can cause catastrophic damage to field crops. Assessment of damage for insurance claims requires insurance inspectors to visit individual fields and estimate damage on individual plants. This study computes temporal profiles and estimates the severity of hail damage to crops in 54 fields through the temporal analysis of vegetation indices calculated from Sentinel-2 images. The damage estimation accuracy of eight vegetative indices in different temporal analyses of delta index (pre-and post-hail differences) or area under curve (AUC) index (time profiles of index affected by hail) was compared. Hail damage was accurately quantified by using the AUC of 32 days of Normalized Difference Vegetation Indices (NDVI), Normalized Difference Water Index (NDWI), and Plant Senescence Radiation Index (PSRI). These metrics were well correlated with ground estimates of hail damage in canola (r = −0.90, RMSE = 8.24), wheat (r = −0.86, RMSE = 12.27), and lentil (r = 0.80, RMSE = 17.41). Thus, the time-series changes in vegetation indices had a good correlation with ground estimates of hail damage which may allow for more accurate assessment of the extent and severity of hail damage to crop land.

The traditional visual rating system is labor-intensive, time-consuming, and prone to human error. Unmanned aerial vehicle (UAV) imagery-based vegetation indices (VI) have potential applications in high-throughput plant phenotyping. The study objective is to determine if UAV imagery provides accurate and consistent estimations of crop injury from herbicide application and its potential as an alternative to visual ratings. The study was conducted at the Kernen Crop Research Farm, University of Saskatchewan in 2016 and 2017. Fababean (Vicia faba L.) crop tolerance to nine herbicide tank mixtures was evaluated with 2 rates distributed in a randomized complete block design (RCBD) with 4 blocks. The trial was imaged using a multispectral camera with a ground sample distance (GSD) of 1.2 cm, one week after the treatment application. Visual ratings of growth reduction and physiological chlorosis were recorded simultaneously with imaging. The optimized soil-adjusted vegetation index (OSAVI) was calculated from the thresholded orthomosaics. The UAV-based vegetation index (OSAVI) produced more precise results compared to visual ratings for both years. The coefficient of variation (CV) of OSAVI was ~1% when compared to 18-43% for the visual ratings. Furthermore, Tukey’s honestly significance difference (HSD) test yielded a more precise mean separation for the UAV-based vegetation index than visual ratings. The significant correlations between OSAVI and the visual ratings from the study suggest that undesirable variability associated with visual assessments can be minimized with the UAV-based approach. UAV-based imagery methods had greater precision than the visual-based ratings for crop herbicide damage. These methods have the potential to replace visual ratings and aid in screening crops for herbicide tolerance.

Legume crops are known to have low soil N uptake early in their life cycle, which can weaken their ability to compete with other species, such as weeds or other crops in intercropping systems. However, there is limited knowledge on the main traits involved in soil N uptake during early growth and for a range of species. The objective of this research was to identify the main traits explaining the variability among legume species in soil N uptake and to study the effect of the soil mineral N supply on the legume strategy for the use of available N sources during early growth. Nine legume species were grown in rhizotrons with or without N supply. Root expansion, shoot and root biomass, nodule establishment, N2 fixation and mineral soil N uptake were measured. A large interspecific variability was observed for all traits affecting soil N uptake. Root lateral expansion and early biomass in relation to seed mass were the major traits influencing soil N uptake regardless of the level of soil N availability. Fenugreek, lentil, alfalfa, and common vetch could be considered weak competitors for soil N due to their low plant biomass and low lateral root expansion. Conversely, peanut, pea, chickpea and soybean had a greater soil N uptake. Faba bean was separated from other species having a higher nodule biomass, a higher N2 fixation and a lower seed reserve depletion. Faba bean was able to simultaneously fix N2 and take up soil N. This work has identified traits of seed mass, shoot and root biomass, root lateral expansion, N2 fixation and seed reserve depletion that allowing classification of legume species regarding their soil N uptake ability during early growth.