next up previous contents
Next: Urban Area Classification via Up: Computational Information Processing Previous: Using External Knowledge in

Tree Level Measurements in Forest Inventory by Digital Camera

Researchers: Jukka Heikkonen, Jouni Juujärvi, Sami Brandt, and Jouko Lampinen

The project is financed by Finnish development center (TEKES) and is carried out with Finnish Forest Research Institute (METLA).

The need for information concerning different natural resources is increasing rapidly. Demands are being made for information which is more accurate, up to date and economical to collect for use in management planning of natural resources, controlling the sustainable use of resources and for monitoring the development of the resources. To be able to carry out these tasks in forestry, it is necessary to extract data from large areas on a regular basis and large amounts of information need to be processed. Traditionally, forest information have been gathered by field sampling methods. However, this technique is expensive and although the data is highly accurate at a point level, its accuracy once generalized often does not meet today's requirements.

Satellite remote sensing is an efficient tool for gathering information from large areas and enables the generalization of field measure information. For example, in the Finnish National Forest Inventory, a multi-source information system has been developed to satisfy new demands for large area information economically and accurately. The information system combines field information with Landsat TM imagery by an application of the k-nearest neighbor method and operationally produces thematic maps detailing the most important forest attributes (such as tree species composition and timber volumes) for the whole of Finland. The traditional field information is gathered from sample plots which are organized in tracts all over the country. In the measurement work, analogue devices such as calipers (for measuring tree diameters) and height meters (for measuring tree heights) are used on sample trees. There are problems with the use of these analogue measurements: they are expensive and the amount of accurate data recorded compared to the time taken to collect the data is not always satisfactory.

In this study, new systems for replacing part of the analogue field measurements are developed by utilizing digital photographs of the sample trees. The methodology presented includes the field proof sample tree photographing system. Automatic pattern recognition methodology is developed for 1) describing the three dimensional space from digital photographs, 2) separating tree stems from the photographs and 3) measuring the interesting variables from the digital image. Accuracy of the variables from the photos is estimated by comparing the results with the results of complete stem analysis. The new digital measurements are proposed for estimating plot level attributes which can be generalized to larger areas by applying the k-nearest neighbor method with Landsat TM satellite images.

Figure 3
Figure 3: Sample image

next up previous contents
Next: Urban Area Classification via Up: Computational Information Processing Previous: Using External Knowledge in
Juha Merimaa