Classification of Horticultural Commodities based on Multispectral UAV Image with k-Nearest Neighbor and Minimum Distance Approach.

Abstract

Horticulture is an agricultural subsector that includes fruit, vegetable, flower and ornamental plants that play an important role in supporting the national economy. Monitoring horticultural cultivation to maintain and improve the quality of horticultural production can utilize remote sensing technology. Remote sensing imagery varies from low resolution such as MODIS (1 km), medium such as Landsat (30 m) and Sentinel-2A (20, 10 m), to high such as IKONOS (1– 4 m). In horticultural agricultural land with a small area, UAV can be used as an alternative because it has high resolution. This study aims to analyze the spectral patterns of horticultural crops based on multispectral UAV images, to map horticultural crops using the k-Nearest Neighbor (k-NN) and Minimum Distance Classification (MDC) methods, and to analyze the level of accuracy of horticultural crop classification using both methods, The research location is at the Pasir Sarongge Experimental Garden, Cianjur Regency using field observation data and multispectral UAV imagery acquired on November 6, 2022. The analysis of the sample spectral patterns was carried out by taking sample areas from 11 classes including horticulture and non-horticulture which were then used to create spectral characteristic curves. Image classification using the k-NN and MDC methods was assessed based on overall accuracy through an error matrixs. The results of the study showed that the spectral response of horticulture was low in visible light where the green band was slightly higher than the red band, while in the red edge band there was a significant increase and continued to increase in the NIR band. The classification of horticultural plants in both methods shows differences in area due to differences in classifier algorithms. The order of the largest to smallest area in k-NN is chili, carrot, potato, banana, cabbage, onion, and tomato. Meanwhile, the order of the largest to smallest in MDC is chili, banana, carrot, cabbage, potato, onion, and tomato. The overall classification accuracy for k-NN and MDC is 89.37% and 51.48% respectively.