Delineating agricultural field boundaries from TM imagery using dyadic wavelet transforms

Agricultural field boundary information is important in the study of agricultural crops by remote sensing. In this paper, the dyadic wavelet transform is adopted to extract field boundaries directly from TM imagery. A field boundary is a set of locally connected pixels characterized by their abrupt spectral intensity variation (singularity) on the image. Preliminary extraction is made on the near-infrared Band 4 (0.76–0.90 μm) image. It is found that most of the boundaries are well characterized by the wavelet modulus maxima curves, particularly for fully vegetated fields, while better results are obtained from the NDVI image for cropped field boundaries. To extract additional boundaries while suppressing unwanted noisy edges, the wavelet transform is then applied to the Wetness image calculated by the Tasselled Cap Transform. A technique is implemented to integrate the edges extracted from each source. The combined results show that the majority of the field boundaries are delineated. The use of a single date image fails to delineate some of the fallow field boundaries and boundaries between two cropped fields with similar spectral properties. Although techniques have been applied to reduce noise, still great difficulties are encountered when trying to delineate edges that are field boundaries from all the detected singularities. Manual editing is still required to trace the remaining boundaries to obtain a complete boundary map. It is recommended that a multi-temporal data set be used to obtain a complete delineation with a fully automated algorithm.