Mapping Potato Plant Density Variation Using Aerial Imagery and Deep Learning Techniques for Precision Agriculture

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