Sadao Fujimura

Nondestructive measurement of chlorophyll pigment content in plant leaves from three-color reflectance and transmittance.

We propose a nondestructive or optical method of measuring the chlorophyll content in a leaf after constructing a mathematical model of reflectance and transmittance of plant leaves as a function of their chlorophyll pigment content. The model is based on the Kubelka-Munk theory and involves the modeling of the multiple reflection of light in a leaf that is assumed to be composed of a stack of four layers. It also includes the assumption that the scattering coefficient and the absorption coefficient of the Kubelka-Munk theory can be expressed as a linear function of the pigment content of a plant leaf. In the proposed method, the chlorophyll content is calculated from reflectances and transmittances at three bands whose center wavelengths are 880,720, and 700 nm. Experiments were performed to confirm the applicability of the model and the method. Reflectance and transmittance calculated with the model showed good agreement with measured values. Furthermore, several unmeasurable constants necessary in the calculation were determined by a least-squares fit. We also confirmed that these results were consistent with several well-known facts in the botanical field. The method proposed here showed a small estimation error of 6.6 microg/cm (2) over the 0-80 microg/cm(2) chlorophyll content range for all kinds of plant tested.

Exterior Algebraic Processing for Remotely Sensed Multispectral and Multitemporal Images

This paper describes remotely sensed multispectral and multitemporal image processing from an algebraic point of view. Especially, image analysis by means of an inner product, an exterior product, and an inner product between two exterior products are presented.

Improvement on depth resolution and reduction of Poisson noise in coded aperture emission CT

We applied a maximum likelihood expectation maximization (ML-EM) method to the reconstruction in coded aperture emission computed tomography (CECT) for the measurement of 3D distribution of a radioactive isotope, where the projection data have a statistical fluctuation obeying a Poisson distribution (Poisson noise). The problem to be solved for CECT is to improve the depth resolution. We made some numerical simulations to confirm that the improvement of depth resolution and the reduction of noise were remarkable. For a point emitter at a distance of 35 cm from the aperture, the depth resolution was about 4 mm even when the average number of photons per pixel in the projection was 10. The quality of images for some 3D objects reconstructed by ML-EM was much better than that by conventional methods.

A binary division algorithm for clustering remotely sensed multi-spectral images

A new method is proposed for clustering remotely sensed multi-spectral images with both high accuracy and high efficiency. For high speed processing, we project image data onto one dimensional sub-space, and limit the number of boundaries in the sub-space. The optimal sub-space and boundary are selected so that the ratio of the variance of within distance to the variance of between distance takes the minimum value. Image data are repeatedly divided into two groups until all of the groups consist of a single cluster. Performance of the proposed method was better than that of ISODATA in both speed and accuracy. The method was successfully applied to actual remotely sensed multi-spectral images. >

Significance-weighted feature extraction from hyperdimensional data

Extracting significant features is essential for processing and transmission of a vast volume of hyper-dimensional data. Conventional ways of extracting features are not always satisfactory for this kind of data in terms of optimality and computation time. Here we present a successive feature extraction method designed for significance-weighted supervised classification. After all the data are orthogonalized and reduced by principal component analysis, a set of appropriate features for prescribed purpose is extracted as linear combinations of the reduced components. The method was applied to 500 dimensional hyperspectral data which were obtained from tree leaves of five categories. Features were successively extracted, and they were found to yield more than several percents higher accuracy for the classification of prescribed classes than a conventional method does.

A method for object-oriented feature extraction from hyperspectral data-generation of new channels by fusion of data

Extracting significant features is essential for processing and transmission of a vast volume of hyperspectral data. Conventional ways of extracting features are not always satisfactory for this kind of data in terms of optimality and computation time. The authors present an object-oriented feature extraction method designed for supervised classification. After all the data are reduced and orthogonalized, a set of appropriate features for the prescribed purpose is extracted as linear combinations (fused channel) of the reduced components. Each dimension of hyperspectral data is weighted and fused according to the extracted features, which means the generation of new channels from hyperspectral data. Results of feature extraction are applied to evaluating the performance of sensors and to designing a new sensor.

Application of an object-oriented feature extraction method to quantitative estimation from hyper-spectral data

Extracting significant features is essential for processing, storing and/or transmission of a vast volume of hyperspectral data. Conventional ways of extracting features are not always satisfactory for this kind of data in terms of optimality and computation time. The authors have already developed an object-oriented feature extraction method designed for supervised classification. They apply the basic idea of the approach to feature extraction for quantitative estimation from hyperspectral data. After the data obtained for various values of a quantity are orthogonalized and reduced by principal component analysis, the features describing the variation of spectra are extracted as linear combinations of the reduced components. An experiment using pigment shows that the feature extraction method for quantitative analysis yielded satisfactory results.

Simultaneous measurement of temperature and absorption of distributed medium by using infrared emission CT

We have developed a method for measurement of temperature and absorption of medium distributed in the three dimensional space. To get accurate temperature distribution of the medium, the absorption of the medium must be known. We achieve this by obtaining the projections of thermal radiation from the medium and an external radiator whose temperature is switched from one to another while obtaining projections. We can reconstruct the distributions from the projections by using an emission CT algorithm. We made an experiment to measure temperature and absorption distribution of flame of a Bunsen burner. The results agreed well with the real temperature profile of the cross section of the flame.<<ETX>>

Method for change analysis with weight of significance using multitemporal multispectral images

A new method is proposed for change analysis with weight of significance between two multi- temporal multi-spectral images. This method gives us areas which indicate the assigned temporal change, for example, from vegetation to bare soil. Image data are projected onto a feature space in which the assigned change is emphasized, and temporal changes between two images are detected with suppression of irrelevant changes. The validity of the method is confirmed by numerical simulation. The method is successfully applied to actual multi- temporal and multi-spectral images.

A Comparison Of Automatic Classification Algorithms For Land Use Map By Remotely Sensed Data

For classification of remotely sensed mu1 tispectral data the most likelihood method is employed very often. It, however, does not always give us the best results. Two reasons for that can be considered, that is, deviation from normal distribution of the data, and lack of generality of the training data used. one of them is dominant is shown. It use:; comparison of correct classification rates obtained by mutually related a1 gori thms. verification is given for the case of classification making land use map. Utilized algorithms were most likelihood, linear discriminant function, minimum Euclidean distance, and correlation coefficient methods. A way to determine which