Object-Based Greenhouse Horticultural Crop Identification from Multi-Temporal Satellite Imagery: A Case Study in Almeria, Spain

Abstract

Greenhouse detection and mapping via remote sensing is a complex task, which has already been addressed in numerous studies. In this research, the innovative goal relies on the identification of greenhouse horticultural crops that were growing under plastic coverings on 30 September 2013. To this end, object-based image analysis (OBIA) and a decision tree classifier (DT) were applied to a set consisting of eight Landsat 8 OLI images collected from May to November 2013. Moreover, a single WorldView-2 satellite image acquired on 30 September 2013, was also used as a data source. In this approach, basic spectral information, textural features and several vegetation indices (VIs) derived from Landsat 8 and WorldView-2 multi-temporal satellite data were computed on previously segmented image objects in order to identify four of the most popular autumn crops cultivated under greenhouse in Almería, Spain (i.e., tomato, pepper, cucumber and aubergine). The best classification accuracy (81.3% overall accuracy) was achieved by using the full set of Landsat 8 time series. These results were considered good in the case of tomato and pepper crops, being significantly worse for cucumber and aubergine. These OPEN ACCESSRemote Sens. 2015, 7 7379 results were hardly improved by adding the information of the WorldView-2 image. The most important information for correct classification of different crops under greenhouses was related to the greenhouse management practices and not the spectral properties of the crops themselves.

AssesSeg—A Command Line Tool to Quantify Image Segmentation Quality: A Test Carried Out in Southern Spain from Satellite Imagery

“…The WV2 data (5 July 2015) and the L8 OLI scene (8 January 2016) images were geometrically and atmospherically corrected using the capabilities of Geomatica v. 2014 (PCI Geomatics, Richmond Hill, ON, Canada). Further details can be found in [12] where a similar pre-processing scheme was undertaken. In particular, the OLI panchromatic band was not used in this test.…”

“…AssesSeg was used to detect the best band combinations for the detection of plastic-covered greenhouses from three multispectral satellite data: Sentinel-2 (S2), Landsat 8 (L8), and WorldView-2 (WV2) images. The authors' choice to test AssesSeg on plastic-covered greenhouses segmentation was due to their sound experience on this active research topic using both pixel-based approaches [5][6][7][8] and OBIA approaches [9][10][11][12][13][14]. The performed test also demonstrates the importance of the number of reference objects in segmentation quality assessment.…”

A Quantitative Assessment of Forest Cover Change in the Moulouya River Watershed (Morocco) by the Integration of a Subpixel-Based and Object-Based Analysis of Landsat Data

“…Moreover, the performance of a textural image varies with the complexity of the study area, the texture measure used, the size of moving window, and the image itself [50]; however, it is widely recognized that textural features are highly suitable for application in very high spatial resolution images [44]. It is worth mentioning that textural information was included in this study by means of standard deviation values, which can be considered as first-order textural features [51]. Mean value for every object computed from the corresponding SMA fraction of all the pixels belonging to the same object [47] Standard deviation Fraction PV, Fraction NPV and Fraction Soil Standard deviation (SD) value for every object computed from the corresponding SMA fraction of all the pixels belonging to the same object [47] …”

Assessment of Multiresolution Segmentation for Extracting Greenhouses From Worldview-2 Imagery

“…The optimal segmentation parameters (300, 0.5 and 0.8 for Scale, Shape and Compactness respectively) were attained by applying a trial-and-error approach. In the other two OBIA studies dealing with greenhouse detection from VHR satellite imagery Aguilar et al, 2015), the segmented objects were generated by manual digitizing, thus avoiding the problems related to segmentation stage (e.g. the setting of multiresolution segmentation parameters).…”