For supervised classification of multispectral images, it is of primary importance to select an appropriate training data set for the categories to be classified. However, as the selection of the training data set is not based on statistical procedures such as random sampling, the estimated distribution parameters for each categories often show biased properties. This paper discusses the correction of the biased estimates for the training data set by the EM algorithm, which is an iterative procedure for obtaining the maximum-likelihood estimates in incomplete data problems. For this purpose, the correction of biased estimates is mathematically formulated as the mixture density problem, where training data and non training data is regarded as the complete data and incomplete data, respectively. In the iterative procedure, the incomplete data are regarded as the pseudo-complete data having the posterior probability (E step), which in turn is utilized to estimate the distribution parameters according to the maximum likelihood method (M step). It is found that the application of the EM algorithm to the multispectral images gives rise to following two problems: (1) inefficiency of the algorithm becomes significant if all pixels in the image are used as the incomplete data, and (2) the results depend on the number of training data selected. The algorithm is modified by introducing the reliability index of the training data to give the stable estimates even if the small number of pixels are used. It is shown that the modified algorithm is successfully applied to the classification of Landsat TM data without losing the good properties of the original EM algorithm.
The relation between luster of a leaf and vegetation and polarization characteristics was investigated through laboratory and field experiment using a polarization analyzer. The following results were obtained. (1) Luster of leaf is influenced by a refractive index and roughness of a leaf facet, thus a lusterless leaf is characterized with larger values of roughness and refractive index compared with a lustered leaf. (2) The dependence of polarization of lusterless leaf on incident angle is smaller than that of the lusterless leaves. There is a tendency that the maximum value of polarization of lusterless leaves appear at a larger incident angle than that of a lustered leaf. (3) Both the polarization and specularity of lustered leaves are larger than those of lusterless leaves, especially at green band. (4) Both an index of polarization(Rq) and a component of specular reflectance(Rs) of vegetation leaves do not depend on wavelength. (5) The informations on polarization is useful for classfication of vegetation.
The environmental database is becoming an indispensable source for various planning processes. However, it is a tough task to keep up the data with the ever changing environment. Especially factors like landuse are rapidly altered in Japan. On the other hand, constant renewal of such man-made data requires a lot of labor and time. Even if we have both, the precision of the data remains dubious because they have to depend on existing information such as maps, aerial photographs and occasional field trips. The integration of LANDSAT TM data with existing enviromental database of Kobe area is tried, and the results are discussed. First, the TM data covering Kobe area is extracted from a scene of 1984. Then band 2, 3 and 4 of the TM data was agreed to the coordinates of exiting database using affine approximation based on 10 corresponding ground control points. Thus agreed TM data was classified into several categories using maximum likelihood method and shown on the display attached to an image processing system named FIVIS. Finally the data was compared with some factors in Kobe environmental database. Strictly speaking, it is impossible to compare the two data because the TM data is based on an instantaneous view of the land surface, while man-made landuse and vegetation data is produced on various data of different stages. Satellite data can supply constant view of the area with reasonable cost though cloud-free data is rare. This is an advantage to follow the changing land surface. However, the view by satellite is often different from landuse we recongnize on vegetation-covered areas and mountain ranges. For instance, paddy fields are major agricultural landuse, but it is perceived as completely different landuses depending on the season by satellite. In addition, classifying landuse or vegetation is difficult due to shadows on the slope. For these reasons, both data should be supplemented each other to obtain more precise view of the area. Then satellite data will contribute a lot to improve current environmental and geographical databases.
Since disaster has a wide variety and complexity in the phenomena, time and space, depending on the disaster category, the natural and artificial conditions of the place of disaster occurrence, the application of remote sensing technology seems to be very effective for disater prevention and management. Therefore, many case studies have been carried out in the various aspect such as flood, landslide, volcanic eruption, earthquake etc. This report summarize the results of many case studies in various disaster categories on the application of remote sensing for detecting the anomalous phenomena for prediction, state of affected area and its damage, and the risk assessment of a disaster. However, low frequency of satellite observation and cloud cover prevent the effective and actual use of satellite data, and furthermore, higher spacial and spectral resolution is required in many cate-gories and quicker data distribution is necessary corresponding to the emergency for disaster prevention. In 1990's, remarkable development of systems for remote sensing are to be expected from the viewpoint of earth observation including global environmental changes due to the expansion of human activities. They will accelerate international cooperation not only on the development but also on intensive effective uses in the field of disaster prevention, especially on the monitoring of global environmental changes to induce anomalous natural phenomena contributing hazardous effects to human life and on technology transfer to developing countries based on IDNDR (International Decade for Natural Disaster Reduction) from 1990. However, to correrpond the actual urgent need for disaster prevention, a new satellite system and its supporting observation systems in the air, sea and on the ground specifically oriented to disaster prevention have to be urgently developed.
Observations of gravity and geomanetic field have recently been made, as a remote-sensing method, for investigations of active fault structure and also for studies of anomalous changes before earthquake occurrences. Anomalies in the distribution of gravity have often turned out to be closely associated with active faults, sometimes with faults which have not been identified by the geomorphological and geological methods. Anomalous changes in gravity were also found in association with crustal uplift. Anomalies in the geomagnetic field have been found in relation to active faults and the anomalies could be interpreted as indicating an anomalous magnetic structure near faults. Such information should be taken into account when observation sites are to be selected for monitoring of anomalous changes before earthquakes, because significant changes are expected to appear near earthquake faults, as demonstrated by a typical example.