Automatic greenhouse delineation from QuickBird and Ikonos satellite images

The area of production under greenhouses has been in rapid growth in recent years, and at present there are over 500,000ha scattered all over the world. Due to the vast amount of inputs (water, fertilizers, fuel, etc.) required, and outputs of various agricultural residues (vegetable waste, plastic sheeting, phytosanitary product containers, etc.), the impact of this type of production system to the environment is considerable in particular if pursued without a sound and sustainable territorial planning. Thus, very high-resolution images provided by satellites such as QuickBird and Ikonos can be useful to plan the expansion of the greenhouse crop-production system in a sustainable way, as well as to reduce the environmental impact of the actual greenhouses improving the inputs use efficiency (water, fertilizers, phytosanitary products, etc.). This work is focused on automated detection of the boundaries of greenhouses using for this purpose the classification results from very high-resolution multi-spectral images of Red, Green, Blue and Near Infrared (RGBNIR) from QuickBird and Ikonos satellites. First, a supervised Maximum Likelihood Classification (MLC) and an Extraction and Classification of Homogeneous Objects (ECHO) both using all the four bands were carried out with focus on plastic greenhouses detection. In the second step, the detected greenhouses according to the classification results were vectorized automatically to generate polygons with irregular borders. In the third step, the irregular contours of polygons were converted into straight lines using an algorithm based on the Hough transformation. The input parameters of this algorithm were obtained by two different procedures: calibration and pseudo-calibration. In the calibration process, the parameter set is obtained from a sample of greenhouses minimizing the area and perimeter errors. In the pseudo-calibration process, a parameter set is obtained for each greenhouse but it is a high time-consumer task. The calibration yielded 66.7% for QuickBird and 49% for Ikonos image of correct delineation of greenhouses. Pseudo-calibration improved success rate and reduced performance difference between the two types of images. The proposed algorithm is able to delineate any greenhouse contour if the proper parameters set is correctly derived.