Fast Extraction Of Plastic Greenhouses Using Worldview-2 Images

By providing controlled areas to produce high quality and wide variety of agricultural products, greenhouses are important for rural development. Their monitoring hence is required for yield estimation and residue management as well as environmental impact. Traditional monitoring approaches based on in situ surveys are costly and time consuming, whereas supervised classification methods require ground truth labels which are often hard to obtain purely. In addition, unsupervised approaches based on approximate spectral clustering produce accurate extraction of (plastic and glass) greenhouses in expense of high computational cost. To extract greenhouses in a fast and unsupervised manner, we employ band thresholding and object based NDVI values, depending on the band characteristics available at Worldview-2 images. Our preliminary experiments produce accuracies as high as those obtained by unsupervised clustering with high computational cost.