Current research projects

12. Automated analysis of forest using high resolution CIR aerial images
    Mats Erikson, Gunilla Borgefors
    Funding: SLU S-faculty
    Period: 9508-
    Partners: Tomas Brandtberg, Dept. of Geology and Geography, West Virginia University, Morgantown, USA; Kenneth Olofsson, Dept. of Forest Resource Management and Geomatics, SLU, Umeå
    Abstract: The main goal of the project is to develop methods for computerised analysis of high spatial resolution remotely sensed data, i.e., digitised aerial photographs and laser scanning data, and to use the results in forestry and environmental assessment instead of (or as a complement to) field visits by humans. A set of 50 research aerial images (digitised colour-IR film), with resolution approximately 10 cm and 3 cm (flight height 600 m, focal length 300 mm) to make the individual tree crowns clearly visible is used. Interesting forest stand parameters to measure in the images are: number and positions of trees, horizontal tree crown areas, tree heights, and tree species composition. Features related to the individual tree species are, e.g., colour, internal structure (texture), and boundary structure.
    During 2001 a tree crown segmentation was developed. This year, this method is being evaluated in a joint project with SLU, Umeå. They have developed another type of method for detecting trees in images. Together we use new (2002) aerial and field data to evaluate both methods against each other and against ground truth. See Figure 8 for a small example of the segmentation results. We have also developed a new segmentation method based on Brownian motion. The method preserves the inner structure of the tree crown in a much better way than the old ones and also finds the contour of the crown in a better way. A method for classification of tree species using, among other clues, the inner structure from the segmentation method based on Brownian motion is in development.
    
    Figure 8: Left: Original tree image. Right: The result from the fuzzy segmentation method together with manually delineated crowns using field data (white polygons).

13. Finding stems in young forests using horizontal laser scanning
    Mats Erikson, Gunilla Borgefors
    Funding: SLU S-faculty
    Period: 0108-
    Partners: Karin Vestlund, Dept. of Forest Products and Markets, SLU
    Abstract: This is a pilot project on finding stems in horizontal laser images in the forest. The goal is to investigate if data can be extracted from such images to automatically select trees for cutting by a thinning robot. The image analysis task is to locate and identify young trees in depth images and also find various features to determine if the tree should be cut or not.

14. Automated segmentation of remotely sensed images over agricultural fields
    Anna Rydberg, Gunilla Borgefors
    Funding: SLU JLT-faculty
    Period: 9602-0112
    Abstract: In this project, multispectral segmentation methods for extraction of remotely sensed features, especially borders, in agricultural fields were developed. An increasing number of satellites, and other sensors, provide more and more information, which creates a need for interactive or even automated analysis of remotely sensed images. The key problem is almost always segmentation. To achieve good segmentation, it is advantageous to integrate several techniques. As the first step in our project, a multispectral edge and line detector was developed. A region growing procedure, which detects additional boundaries, followed this step. This results in an over-segmented image, therefore segmentation was followed two merging procedures, where one procedure merged small regions with similar spectral characteristics, and one procedure merged regions according to their shape. A priori knowledge about field shape and size should be taken into account, if possible. This project resulted in Rydberg's PhD thesis, defended in November 2001. This year, a book chapter the book ``Geospatial Pattern Recognition'', describing the whole process, was published.

15. New techniques for information extraction from hyperspectral crop reflectance data
    Hamed Hamid Muhammed
    Funding: UU TN-faculty, Swedish National Space Board
    Period: 0201-
    Partners: Anders Larsolle, Dept. of Agricultural Engineering, SLU
    Abstract: Hyperspectral crop reflectance data can be used for studying the pathological condition of the crop. The influence of pathological status of a crop on its spectral characteristics is detectable in the visible and/or the near-infrared regions of the electromagnetic spectrum. FVBA can be used here to analyse and study the pathological condition of a crop, by taking a reference or training data set consisting of hyperspectral data vectors and the corresponding field measurements of the leaf-damage level in the studied crop. Then, the damage levels can be estimated for new collected hyperspectral data vectors. It has been noticed that differences in the spectral characteristics between normal or healthy crops and others suffering from physiological stress or disease, can be revealed and/or magnified by simply normalising the data properly. Such effects can be achieved when normalising each hyperspectral reflectance data vector into a zero-mean and unit variance vector (i.e., a whitened data vector is obtained), and then performing a band-wise normalisation on the previous results (i.e., putting all elements at a certain wavelength interval or band in one vector and whitening it). Using these normalised vectors in FVBA gives better results. Also, it has been noticed that using normalised hyperspectral data, including the training data, gives good results when a simple nearest neighbour classifier is used to classify our data against the training data. The correlation coefficient and the sum of squared differences are used as distance measures in the nearest neighbour classifier. High correlation is obtained, between the results (of using FVBA and the nearest neighbour classifier) and the corresponding field measurements, confirming the usefulness and efficiency of these methods for this type of analysis.

16. Digital video and colour camera in remote sensing of water
    Tommy Lindell
    Period: 0001-
    Partners: CNR, Milan, Italy
    Abstract: Test of the usefulness of air-borne digital camera and video for mapping water variables. Lindell has been constructing a holder for the digital video/camera for small aircraft. Data have been collected from Lakes Erken and Mälaren, and from coral bottoms in Bisceyne National Park.
    Recently, tests of the usefulness of those images have been performed for the classification of the Swedish coastline.

17. Detecting Coral Reef Bleaching from Optical Satellites: a pilot and demonstration project (CORBOS)
    Petra Philipson, Tommy Lindell
    Funding: Foundation for Strategic Environmental Research (MISTRA), RESE programme
    Period: 0001-0212
    Partners: Swedish Meteorological and Hydrological Institute (SMHI), Norrköping
    Abstract: Recent dramatic bleaching events on coral reefs have enhanced the need for global environmental monitoring. The development of remote sensing methods for monitoring of coral reefs requires investigation of the sensor limitations, the optical properties of the bottom features and understanding of the influence of the atmosphere and water column on the collected remote sensing data. This project have been investigating the possibilities of using remote sensing technique for coral reef monitoring and change detection, with focus on detection of coral bleaching using existing satellite sensors. We have compared an IRS LISS-III image taken during the 1998 bleaching event in Belize to images taken before the bleaching event. The sensitivity of different sensors has been investigated and a simulation has been made to estimate the effect of sub-pixel changes. A manual interpretation of coral bleaching, based on differences between the images, was performed and the outcome has been compared to field observations. The spectral characteristics of the pixels corresponding to the field observations and the manually interpreted bleaching have been analysed and compared to pixels from unaffected areas, with positive results for the detection of bleaching from medium resolution satellites. Procedures for an automated analysis have been tried to make monitoring more efficient.
    A field study has been performed in Belize in 2002 and together with the use of SPOT and IKONOS images further improvements have been achieved in detecting changes on coral bottoms.

18. HYSENS -- Hyperspectral remote sensing using a new version of ROSIS
    Tommy Lindell, Petra Philipson
    Funding: ESA/DLR
    Period: 0001-
    Partners: Don Pierson, Dept. of Evolutionary Biology, Limnology, UU; Eugenio Zilioli, CNR, Milan, Province Environment Protection Agency of Trent (APPA), Province Ecological Agency of Verone (ECOV) and Regional Environment Protection Agency, Verone, all in Italy
    Abstract: ROSIS for Algal Mapping in Lacustrine Environment (ROSALMA). Rosalma is essentially oriented to a double task:
    1. to correlate basic water quality parameters like chlorophyll, suspended sediment concentrations and Secchi disc to the hyperspectral data by using a semi-analytical approach already proved in other geographic conditions and with other hyperspectral devices;
    2. to determine the best optical spectral windows for mapping the macrophyte growth, in order to design a possible operational tool to be used for environmental emergencies of this kind, especially in mapping its spatial distribution.
    Lindell & Philipson have participated in the work on Lake Garda, Italy earlier and in the evaluations.
    The focus of the last part of this project will be devoted to applications of the MERIS sensor to water quality monitoring, using the experiences gained from the CASI and ROSIS sensors.

19. Industrial plume detection in multispectral remote sensing data
    Petra Philipson, Tommy Lindell
    Funding: Foundation for Strategic Environmental Research (MISTRA), RESE programme
    Period: 0001-0212
    Partners: Marcus Liljeberg, IVL - Swedish Environmental Research Institute, Stockholm;
    Niklas Strömbeck, Dept. of Evolutionary Biology, Limnology, UU
    Abstract: There are a number of paper and pulp industries located along the East coast of Sweden. The amount of substances discharged into the coastal sea water is regulated for each industry, but the size of the area affected by each outlet is relatively unknown and varies during the year.
    The general objective for this project was to investigate if and to what degree remote sensing data could be used to locate and map the extent of the industrial plumes. Such an investigation involves the analysis of the optical properties of plumes in comparison to the properties of natural water constituents. The atmospheric influence on the remote sensing data must also be considered for any aquatic application that should result in general descriptions of the properties and quantitative estimations of the substances present in the water.
    Airborne hyperspectral images and laboratory measurements of water samples have been used to investigate if there are any spectral properties related to paper mill discharges that can be useful for identification of the existence and concentration level of the discharges using present and future remote sensing data.
    Besides correlation studies between remote sensing data and field data, the spectral properties of the discharge water have been investigated and used in the analysis of available remote sensing data.

20. Remote sensing for change detection and monitoring of Case II and lake waters
    Petra Philipson, Tommy Lindell
    Funding: Foundation for Strategic Environmental Research (MISTRA), RESE programme
    Period: 9701-0212
    Partners: Swedish Meteorological and Hydrological Institute (SMHI), Norrköping
    Abstract: The ability to map and monitor water quality parameters in Case II and lake waters is of great environmental interest. Images from spectrographic sensors constitute an important part of such a mapping and monitoring system. The Compact Airborne Spectrographic Imager (CASI) was used to collect images over the archipelago of Stockholm, Lake Mälaren and Lake Erken in August 1997. These images have been evaluated in combination with simultaneously collected field data. The work in the archipelago has been concentrated on finding relations between the water quality variables and the reflectance measurements from the field and correlation analysis between field and scanned data (CASI). It is unlikely, though, that the resulting algorithms from these kind of empirical relationships will be sufficiently general to be used in a variety of contexts. In recent years, the focus of lake water monitoring by remote sensing, has shifted towards coupling remotely sensed data to semi-analytical modelling. A simple bio-optical model for the water environment in lakes has been developed. The model is mainly based on oceanographic relationships from the literature. A large historical data set of water quality measurements have been used together with the model to develop algorithms for the retrieval of water quality parameters. The model takes as inputs the concentration of chlorophyll, the concentration of suspended particulate inorganic material (SPIM) and the absorption of coloured dissolved organic matter (CDOM) at 420 nm. The output from the model is a reflectance spectrum just above the water surface. From the modelled reflectance, algorithms are derived for chlorophyll, SPIM and CDOM absorption at 420 nm. The algorithms were applied to the atmospherically corrected CASI data from Lake Mälaren and Lake Erken, see Figure 9. The resulting concentration maps were validated using ground truth measurements. The results from the validation of the CASI algorithms are satisfying, and the modelled concentrations and absorption coefficients corresponds well to the ground truth measurements, which is very encouraging for the future work. The work has been presented in reviewed international publications.
    
    Figure 9: Chlorophyll concentrations in Lake Erken on the 6th of August, 1997. The estimation is made by applying algorithms derived from the bio-optical modelling to airborne remote sensing data.

21. Acquisition of hyperspectral data under the ocean surface
    Julia Åhlén, Tommy Lindell, Ewert Bengtsson
    Funding: Dept. of Mathematics, Natural Sciences, and Computing, University College of Gävle; The KK-foundation
    Period: 0102-
    Abstract: The examination of image processing techniques for dealing with image enhancement in underwater conditions is important for scientists involved with marine environments. One application could be a study of archaeological sites in various oceans of the world. Generally, historical objects found under the water have to be analysed directly with photography. Another application is a different approach to study problems observed on corals such as bleaching. Prominent blue colour of clear ocean water, apart from sky reflection, is due to selective absorption by water molecules. Due to this nature of underwater optics, red light diminishes when the depth increases, thus producing blue to grey like images. In fact all red light is gone when reaching 3 m of depth. So far very few studies have been done on multi- or hyperspectral data taken under the water. Such studies could develop techniques to efficiently reduce the negative effects of scatter and light absorption. These effects often result in bluish images. In this project we are investigating how multi- or hyperspectral data can be utilised to give us better colour information in underwater images. Initially we are looking at what techniques are available for creating an image acquisition system that could give multi- or hyperspectral data. Approaching the issues of enhancement for underwater images from the perspective of colour constancy is one of the approaches that are being investigated. In a cooperation with the University of Southern Florida, St. Petersburg a field session was performed in the spring of 2002 in the Mexican Gulf.

22. New techniques for information extraction from remotely sensed hyperspectral images
    Hamed Hamid Muhammed, Tommy Lindell, Ewert Bengtsson
    Funding: UU TN-faculty, Swedish National Space Board
    Period: 0001-
    Abstract: A substance can be characterised and recognised by its spectral signature. The benefit of hyperspectral imagery is that a sufficient number of narrow spectral bands is available to be able to accurately determine the spectral response at each pixel in the image. A pixel (or a point spectrum) in a hyperspectral image can be considered as a mixture of the reflectance spectra of several substances that can be found in the (remotely sensed) imaged region. The mixture coefficients correspond to the (relative) amounts of these substances in the studied region. Linear transformation methods can be used to project the hyperspectral data on the basis vectors found by the used transformation. Independent Component Analysis (ICA) and Principal Component Analysis (PCA) have been used to transform the hyperspectral data as a first step to get a new set of data that is more suited for further processing than the original data. The next step is to interpret and use the ICA or PCA results efficiently. This can be achieved by using a new technique called Feature-Vector Based Analysis (FVBA) which has been developed during 2001. The outputs of the transformation step (which are a number of basis vectors and projections of the original data on these vectors) are considered as so called Component-FeatureVector pairs in the subsequent FVBA step. The FVBA task itself is application dependent. But, the common idea of FVBA is to look at the Feature Vectors to understand the corresponding Components. FVBA can be used for four main types of applications. Two of them can directly be distinguished where either well-defined Feature Vectors or well-defined Components are obtained. The other two types of applications are feature extraction and classification. When studying hyperspectral images, the obtained Feature Vectors and the corresponding Components represent the spectral signatures and the corresponding weight coefficients images (the relative concentration maps) of the different constituting substances. During 2001, the work has resulted in two publication at reviewed conferences, one describes FVBA itself, and another one about using FVBA for analysing hyperspectral images.