Potential of low altitude multispectral imaging for in-field apple tree nursery inventory mapping

Aerial imaging based plant counting is a widely studied topic in remote sensing. Nevertheless, existing methodologies are not applicable for in-field nurseries tree counting due to the complexity in canopy shapes, irregular plant spacing and growth, and diverse textures captured on the images. In this study, a new algorithm has been developed in accordance to the specific requirements of apple nurseries images. Algorithm composed of two key steps, i.e. a raster processing followed by a vector analyses. First step attempt to isolate apple plant pixels using spatial, spectral, and radiometric enhancements. In vector processing, filters based on the size and location of the polygons were applied to isolate the areas resulting from earlier step to represent apple plants. Algorithm was evaluated to estimate number of apple trees in an young nursery imaged with low altitude multispectral imaging system at four altitudes of 10, 25, 40, and 50 m. Multispectral imaging sensor consisted of near-infrared (NIR), green and blue as three bands. For 10- and 25-m images, algorithm performance was evaluated in individual as well as in mosaic images. Low altitude images with ≤ 25 m above ground level were ideally suited for young apple nursery tree count with 5% or less estimation error. Tree count accuracy was 97% and 95% for 10-m altitude individual and mosaic images, respectively. Similarly, those values were 92% and 88% for 25-m altitude images. Based on the results, images at 40 m are recommended only when the methodology include four extra steps added to the base algorithm to have tree count accuracies of about 88%. Images at 50 m are not recommended in any case due to the low accuracy obtained (~ 75%). Overall, the low altitude multispectral imaging integrated with image processing algorithm developed in this study will aid nursery growers in low-cost and timely tree count needed for inventory mapping and management.