Extracting banana farm parameters from NIR images

In this paper, a near-infrared image of a banana plantation was taken from an unmanned aerial vehicle using a multispectral camera. The image was processed to extract banana plantation farm parameters such as number of banana plants present and classifying the sizes as small, medium, and large. The image was preprocessed using contrast-stretching and median filtering to enhance the image and remove noise. Top-hat filtering was used to approximate the background, which was then subtracted from the main image to get the pixels that contain banana plants. A binary image was created using a global threshold with banana plants as foreground. The binary image was cleaned using image opening and individual banana plants were extracted using connected components via 8-connectivity and were classified by size by computing the area of the connected components. It was shown that majority of the bananas were extracted and classified but there were also misclassification such as non-banana components that were included and banana plants with overlapping leaves, which appear as one component having a large area.