Object Based Forest Stand Density Estimation from Very High Resolution Optical Imagery Using Wavelet Based Texture Measures

Stand density or tree density, expressed as the number of trees per unit area, is an important forest management parameter. It is used by foresters to evaluate regeneration, to assess the effect of forest management measures or as an indicator variable for other stand parameters like age, basal area and volume. A stand density estimation technique, based on the application of wavelet analysis, is presented in this work. First, very high resolution imagery is segmented using a wavelet based segmentation algorithm. Then, wavelet coefficient statistics are calculated for each separate object in order to characterize local texture. Next, the wavelet statistics are related to stand density using an artificial neural network. The established relation is finally used to produce semi-continuous stand density maps. In order to test the proposed method, several color-infrared aerial photographs (scale 1:5000) were scanned, mosaiced and converted to 25cm resolution orthophotos. A large number of trees (45,000) were manually digitized on this dataset. The available imagery was degraded to 1m and 2m spatial resolutions, and the new algorithm was applied. To put the presented method in perspective, a comparison was made with the local maximum filter, a conventional density estimation and individual tree identification method. Analysis of variance revealed that the wavelet based technique performed significantly better (p < 0.001) when compared with the local maximum based method, in terms of correlation, root mean square error and mean absolute difference, regardless of resolution or spectral band.