Ear density is one of the yield components of wheat and therefore a variable of high agronomic interest. Its traditional measurement necessitates laborious human observations in the field or destructive sampling. In the recent years, deep learning based on RGB images has been identified as a low-cost, robust and high-throughput alternative to measure this variable. However, most of the studies were limited to the computer challenge of counting the ears in the images, without aiming to convert those counts into an ear density, i.e. a number of ears per square metre in the field. Moreover, we suspect that ear detection performance indicators based on image labelling may be misleading: if direct sunlight induces a shadow that hide an ear for both the human image annotator and the deep learning model, no error will be detected while the ear density will be underestimated. Consequently, the aim of this study was not only to propose a method for automatic measurement of ear density, but also to evaluate the potential impact of the sun on the measurement.

A same zone of a wheat plot has been imaged by two nadir RGB cameras all over the daily course of the sun, this repeated at four key wheat development stages. The bounding boxes of the ears were detected using the YOLOv5 deep learning model, trained on rich existing wheat ear datasets. The shifts between the same elements observed in the images from the two cameras were exploited to compute the image footprint by stereovision. The ear count, divided by the image footprint, yielded the ear density. To investigate the effect of the sun, a solar spectrum was recorded thanks to a spectrometer at the time of each image acquisition.

The real image footprint did not change during a day, but the image footprint estimated by stereovision showed a standard deviation of 0.009 m² on the very cloudy day and of 0.015 m² on the day that was cloudy most of the time. The two other days, with alternation of clouds and sun, standard deviations were 0.029 and 0.053 m². It seems that the sun impacted the stereovision performances, although other factors such as the wheat development stage, could also had an influence. An error of 0.053 m² on the footprint was translated by an error on ear density of around 15 ears/m², which is acceptable regarding the ear densities superior to 300 ears/m². A constant footprint was selected to investigate the other sources of variation of ear density. On the very cloudy day, the ear density was roughly constant, but on other days, lower ear densities were measured when the sun irradiance increased. That error could be up to 50 ears/m². The study of the human image labels will tell us if the human annotators also identify less ears in sunny images. Ear detection performances will also be computed and studied at the four wheat development stages all along the day.