Optimal Selection of Locations for Wheat Ear Smartphone Sampling Using UAV Multispectral Data

This research aims to improve the smartphone image acquisition procedure for wheat yield prediction involving the non-uniformity of wheat fields in the final yield prediction. We compare two image acquisition methods—random smartphone sampling and a proposed optimized approach—using UAV multispectral data. The data was obtained from a DJI Phantom 4 Multispectral, collected in mid-May and June from two wheat fields in Vojvodina: Kovilj and Šajkaš. The Normalized Difference Vegetation Index and Soil-Adjusted Vegetation Index were used to segment fields into four management zones. The median value of the Soil-Adjusted Vegetation Index for each zone and the nearest location in terms of distance from the median point represent optimal locations. At these locations, wheat ears were sampled to estimate yield. The results demonstrate that the optimized approach provides a more efficient and reliable method, leading to significant time savings and ensuring precise early wheat yield predictions. At the optimal locations in both fields, the yield estimates are consistently underestimated compared to the actual yield, while at the randomly selected locations, the yield is generally overestimated, with underestimation occurring in only one case. The strong correlation between the Soil-Adjusted Vegetation Index and wheat yield, averaging 0.89 (0.88 for Kovilj and 0.90 for Šajkaš), underscoring the effectiveness of the proposed method in identifying areas with greater productivity potential.