In potato (Solanum tuberosum) production, the number of tubers harvested and their sizes are related to the plant population. Field maps of the spatial variation in plant density can therefore provide a decision support tool for spatially variable harvest timing to optimize tuber sizes by allowing densely populated management zones more tuber-bulking time. Computer vision has been proposed to enumerate plant numbers using images from unmanned aerial vehicles (UAV) but inaccurate predictions in images of merged canopies remains a challenge. Some research has been done on individual potato plant bounding box prediction but there is currently no information on the spatial structure of plant density that these models may reveal and its relationship with potato yield quality attributes. In this study, the Faster Region-based Convolutional Neural Network (FRCNN) framework was used to produce a plant detection model and estimate plant densities across a UAV orthomosaic. Using aerial images of 2 mm ground sampling distance (GSD) collected from potatoes at 40 days after planting, the FRCNN model was trained to an average precision (aP) of 0.78 on unseen testing data. The model was then used to generate predictions on quadrants imposed on orthorectified rasters captured at 14 and 18 days after emergence. After spatially interpolating the plant densities, the resultant surfaces were highly correlated to manually-determined plant density (R2 = 0.80). Further correlations were observed with tuber number (r = 0.54 at Butter Hill; r = 0.53 at Horse Foxhole), marketable tuber weight per plant (r = −0.57 at Buttery Hill; r = −0.56 at Horse Foxhole) and the normalized difference vegetation index (r = 0.61). These results show that accurate two-dimensional maps of plant density can be constructed from UAV imagery with high correlation to important yield components, despite the loss of accuracy of FRCNN models in partially merged canopies.

Accurate estimation of tuber size distribution (TSD) parameters in discretely categorized potato (Solanum tuberosum L) yield samples is desirable for estimating modal tuber sizes, which is fundamental to yield prediction. In the current work, systematic yield digs were conducted on five commercial fields (N = 119) to compare the Weibull, Gamma and Gaussian distribution functions for relative-likelihood-based goodness-of-fit to the observed discrete distributions. Parameters were estimated using maximum likelihood estimation (MLE) for the three distributions but were also derived using the percentiles approach for the Weibull distribution to compare accuracy against the MLE approaches. The relationship between TSD and soil nutrient variability was examined using the best-fitting model's parameters. The percentiles approach had lower overall relative likelihood than the MLE approaches across five locations, but had consistently lower Root Mean Square Error in the marketable tuber size range. Negative relationships were observed between the percentile-based shape parameter and the concentrations of Phosphorus and Nitrogen, with significant (non-zero-overlapping 95% confidence interval) regression coefficients for P (−0.74 ± 0.33 for distribution of proportional tuber numbers and −1.3 ± 0.62 for tuber weights). Stem density was negatively associated with the scale and mode of tuber number (regression coefficients −0.98 ± 0.63 and −1.08 ± 0.78 respectively) and tuber weight (regression coefficients −0.99 ± 0.78 and −1.04 ± 0.69 respectively) distributions. Phosphorus is negatively related to the scale of the tuber-number-based distribution while positively associating with the tuber weight distribution. The results suggest that excess P application was associated with the increase in small tubers that did not contribute significant weight to the final yield.

Satellite Image Time Series (SITS) have been used to build models for predicting Potato (Solanum tuberosum L.) yields at regional scales, but evidence of extension of such models to local field scale for practical use in precision agriculture is lacking. In this study, multispectral data from the Sentinel-2 satellite were used to interpolate continuous spectral signatures of potato canopies and generate vegetation indices and the red edge inflection point (REIP) to relate to marketable yield and stem density. The SITS data were collected from 94 sampling locations across five potato fields in England, United Kingdom. The sampling locations were georeferenced and the number of stems per square meter, as well as marketable yield, were determined at harvest. The first principal components of the temporal variation of each SITS wavelength were extracted and used to generate 54 vegetation indices to relate to the response variables. Marketable yield was negatively related to the overall seasonal reflectance (first principal component) at 559 nm with a beta coefficient of −0.53 (±0.18 at p = 0.05). Seasonal reflectance at 703 nm had a positive significant relationship with Marketable yield. Marketable yield was modeled with a normalized root mean square error (nRMSE) of 0.16 and R2 of 0.65. On the other hand, Stem density was significantly related to the Specific Leaf Area Vegetation Index (β = 1.66 ± 1.59) but the REIP’s farthest position during the season was reached later in dense canopies (β = 1.18 ± 0.79) with a higher reflectance (β = 3.43 ± 1.9). This suggested that denser canopies took longer to reach their maximum chlorophyll intensity and the intensity was lower than in sparse canopies. Potato stem density was modeled with an nRMSE of 0.24 and R2 of 0.51. These results reinforce the importance of SITS analysis as opposed to the use of single-instance intrinsic indices.

Applying colourbased feature extraction and transfer learning to develop a high throughput inference system for potato (Solanum tuberosum L.) stems with images from unmanned aerial vehicles after canopy consolidation', Precision Agriculture.

No abstract