358
registers and checking of farmers’ aid
applications.
Existing research on tree crown identification is
often still experimental (e.g. Larsen, 1997) and
usually species-related. For olive trees, two
operational - but proprietary - “tree counting”
algorithms were identified, but these were not
directly suitable (for technical reasons) nor
public domain.
The authors have developed a prototype tool -
named Olicount - for the computer-assisted
counting, on scanned aerial photography, of
olive trees. The software - developed in C++ on
a Microsoft Windows 95™ platform - has been
publicly available since mid-1997. Olicount
uses a mix of techniques (image thresholding,
region growing, tree parameter testing) to
produce a map of candidate objects: in the GIS
environment these objects are verified using
classical photo-interpretation and a final count
made.
The project itself was broken down into four
steps:
i. Collection of existing know-how on
automated or computer-assisted counting of
(olive) trees on digitised aerial photography;
i. Development of an algorithm, starting from
the information collected;
ii. Implementation of the algorithm and
integration within an existing GIS application
environment
iv. Distribution of the knowledge, algorithm and
software tools to interested organisations
Step 1 included contact with commercial
companies already operating semi-automatic
counting systems for Olive trees in Europe.
All phases of the project are now complete, and
the tool has entered into trial or operational use
with a number of organisations, including the
JRC, the Italian Ministry of Agriculture (AIMA),
and the French administration responsible for
olive oil subsidy management (SIDO, 1998).
1.3 Image data; suitability, availability
The counting of trees is a classic forestry
application for remote sensing data. Aerial
photography is normally considered the most
suitable, due to characteristics of spatial
resolution. Nevertheless, crown counting can
be difficult and inaccurate; much depends on
the quality of the image data, the physiognomy
of the stand, and the skill and experience of the
photointerpreter. Howard (1991) cites a
minimum spacing of +/-4m between trees for
large scale aerial photography, and that most
accurate counts can be obtained in boreal
forest, open grown woodlands and recently
thinned plantations. Olive trees stand
conditions can be considered - in the majority
of cases - to conform to these conditions.
Most forest inventory approaches for the
estimation of tree numbers (or their derivative
products, such as timber volume) rely on
statistical sampling, not individual tree
counting. But the regulatory context, parcel-
level requirement for data, and complex
agronomic characteristics (irregular, mixed,
etc.) of olive tree stands (plantations) demand
an object-identification approach.
The approach adopted here is essentially
morphometric, and spatial resolution is
therefore of highest importance in the image
recognition of the trees. Productive - i.e. 5 to
100 year old - olive trees typically have crowns
of 3m to 12m in diameter, with a spacing of 6m
to 10m. In this project and work by associated
Member State administrations (e.g., SIDO
1998), 1m pixel-size imagery has proven to be
sufficient in most contexts of olive tree
production.
From a radiometric perspective, internal
investigations have not yet determined a
significant advantage of multi-spectral data or
colour emulsions over standard panchromatic
film. Of significance at a project management
level, however, is data volume, and single
band, 8-bit data helps to keep this down.
Nevertheless, the gross area covered by olive
trees in Europe represents around 1 Tera-byte
of 8-bit, 1m pixel data. Viewed from another
perspective, an olive tree can be represented in
such a image with just 100 bytes (10x10
pixels), and still remain interpretable - an
efficient and compact description in any terms.
Suitable data, therefore, for this operation is
scanned panchromatic aerial photography, with
a pixel size of 0.8m-1m, typically derived from
high resolution 1:40,000 scale flights. In
connection with other EU-related agricultural
subsidy projects, such digital data are available
for the whole of Italy and Portugal (ortho-
imagery), or have been acquired for the
projects concerned (Spain, France, Greece).
This was also an important issue in deciding to
use this type of data for prototype model
development.
2. Overview of algorithm
2.1 The problem
Two problems characterise the identification of
olive trees in the context of this work.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
Br
