• Proceedings of the Korean Society of Near Infrared Spectroscopy Conference
• /
• 2001.06a
• /
• pp.1154-1154
• /
• 2001

Near infra-red (NIR) spectroscopy has been used for the non-invasive assessment of intact fruit for eating quality attributes such as total soluble solids (TSS) content. However, little information is available in the literature with respect to the robustness of such calibration models validated against independent populations (however, see Peiris et al. 1998 and Guthrie et al. 1998). Many studies report ‘prediction’ statistics in which the calibration and prediction sets are subsets of the same population (e. g. a three year calibration validated against a set from the same population, Peiris et al. 1998; calibration and validation subsets of the same initial population, Guthrie and Walsh 1997 and McGlone and Kawano 1998). In this study, a calibration was developed across 84 melon fruit (R$^2$= 0.86$^{\circ}$Brix, SECV = 0.38$^{\circ}$Brix), which predicted well on fruit excluded from the calibration set but taken from the same population (n = 24, SEP = 0.38$^{\circ}$Brix with 0.1$^{\circ}$Brix bias), relative to an independent group (same variety and farm but different harvest date) (n = 24, SEP= 0.66$^{\circ}$ Brix with 0.1$^{\circ}$Brix bias). Prediction on a different variety, different growing district and time was worse (n = 24, SEP = 1.2$^{\circ}$Brix with 0.9$^{\circ}$Brix bias). Using an ‘in-line’ unit based on a silicon diode array spectrometer, as described in Walsh et al. (2000), we collected spectra from fruit populations covering different varieties, growing districts and time. The calibration procedure was optimized in terms of spectral window, derivative function and scatter correction. Performance of a calibration across new populations of fruit (different varieties, growing districts and harvest date) is reported. Various calibration sample selection techniques (primarily based on Mahalanobis distances), were trialled to structure the calibration population to improve robustness of prediction on independent sets. Optimization of calibration population structure (using the ISI protocols of neighbourhood and global distances) resulted in the elimination of over 50% of the initial data set. The use of the ISI Local Calibration routine was also investigated.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.30 no.4
• /
• pp.397-404
• /
• 2012

Jeju where concentrates on agriculture and tourism, conversion of outdoor culture into cultivation under structure happens actively for the purpose of increasing profit so continuous examination on house cultivation area is very important for this region. This paper is to suggest the effective image classification method using high resolution satellite image to detect the greenhouse. We carried out classification of greenhouse using the supervised classification and rule-based classification method about Formosat-2 images. Connecting result of two classification try to find accuracy improvement for greenhouse detection. Results about each classification method were calculated the accuracy by comparing with the result of visual detection. As a result, mahalanobis distance among the supervised methods was resulted in the highest detection. Also, it could be checked that detection accuracy was improved by tying with result of supervised method and result of rule-based classification. Therefore, it was expected that effective detection of greenhouse would be feasible if henceforward further study is performed in the process of connecting supervised classification and rule-based classification.

• Journal of Korea Multimedia Society
• /
• v.14 no.10
• /
• pp.1311-1322
• /
• 2011

Gesture is applied in many applications such as smart-phone, tablet-PC, and web-browser since it is a fast and simple way to invoke commands. For gesture applications, a gesture designer needs to consider both user and system during designing gestures. In spite of development of gesture design tools, some difficulties for gesture design still remains as followings; first, a designer must design every gesture manually one by one, and, second, a designer must repeatedly train gestures. In this paper, we propose a gesture proposal system that automates gesture training and gesture generation to provide more simple gesture design environment. Using automation of gesture training, a designer does not need to manually train gestures. Proposed gesture proposal system would decrease difficulties of gesture design by suggesting gestures of high recognition possibility that are generated based on mahalanobis distance calculation among generated and pre-existing gestures.

• Korean Journal of Remote Sensing
• /
• v.30 no.2
• /
• pp.207-218
• /
• 2014

The information on the land cover along stream corridor is important for stream restoration and maintenance activities. This study aims to review the different classification methods for mapping the status of stream corridors in Seom River using airborne RGB and CIR digital ortho imagery with a ground pixel resolution of 0.2m. The maximum likelihood classification, minimum distance classification, parallelepiped classification, mahalanobis distance classification algorithms were performed with regard to the improvement methods, the skewed data for training classifiers and filtering technique. From these results follows that, in aerial image classification, Maximum likelihood classification gave results the highest classification accuracy and the CIR image showed comparatively high precision.

• Journal of the Korean Society of Tobacco Science
• /
• v.16 no.2
• /
• pp.163-171
• /
• 1994

This experiment was conducted to investigate the discrimination of cigarette brands and the similarity between Korea and America cigarette brands by near infrared spectra. Statistical tools such as principal component analysis (PCA) and mahalanobis distance(M.D) were used. The discrimination rate of the Korea and the America cigarette brands, determined by position number which was calculated with PCA and M.D, was 94% and 87%, respectively. The spectra of the 10 America cigarette brands were selected by averaging 5 sample spectra for each brand and another 5 spectra for each brand were investigated to use as the sample spectra. Comparing the sample spectra with the selected spectra by M.D using 410-1090 nm, 1110-1850 nm and 1970-2490 nm wavelength, the discrimination rate which was determined by the closest M.D between the sample and the selected spectra was 100% when each spectra was investigated on the same time. But the discrimination rate decreased 50% when the sample and the selected spectra were investigated on the different time. Excluding 1970-2490 nm wavelength, the discrimination rate increased up to 90% when the sample and the selected spectra were investigated on the different time. Comparing the spectra of Korea cigarette brands with those of America cigarette brands by M.D using only 410-1090 nm and 1110-1850 nm wavelength, the spectra of Expo(G) was similar to Winston, Vantage(U.L) and Benson & hedges(M.), the spectra of Hanaro(D) was similar to Carrel, Winston(L), Vantage(U.L), \Vantage and Carrel(L), the spectra of Hanaro(L) was similar to Winston(L) , Carton, Vantage and Carmel(L) and the spectra of Pinetree was similar to Kent, Kool, Kent(G.L), Merit and Benson & hedges(L), respectively. Key Words : NIRS, cigarette brands discrimination

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.6
• /
• pp.783-788
• /
• 2004

Many reported methods assume that the faces in an image or an image sequence have been identified and localization. Face detection from image is a challenging task because of the variability in scale, location, orientation and pose. The difficulties in visual detection and recognition are caused by the variations in viewpoint, viewing distance, illumination. In this paper, we present an efficient linear discriminant for multi-view face detection and face location. We define the training data by using the Fisher`s linear discriminant in an efficient learning method. Face detection is very difficult because it is influenced by the poses of the human face and changes in illumination. This idea can solve the multi-view and scale face detection problems. In this paper, we extract the face using the Fisher`s linear discriminant that has hierarchical models invariant size and background. The purpose of this paper is to classify face and non-face for efficient Fisher`s linear discriminant.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.35 no.2
• /
• pp.126-131
• /
• 1990

Sixty two Job's Tear cultivars were largely classified into ten varietal groups and every varieties. except for Group V forming the Group by one variety. uniformly formed the Groups. From Group I to Group Ⅹ respectively contained three(5%), eighteen(29%), five(8%), thirteen(21%), one(2%), five(8%), seven(11%), four(7%), two(3%) and four(7%) varieties. Group II and Group IV showed considerably large variation whereas Group Ⅷ, IV and Ⅹ showed low variation and inferiority in vigorosity and yield components. Most of the varietal Group were not associated with their geographical origin. Days to flowering and plant height among the nine characters were the largest contributors to the D$^2$ in both inter- and inter groups.

• Journal of Fashion Business
• /
• v.11 no.2
• /
• pp.58-75
• /
• 2007

This study held human measurement for 200 male subjects in their twenties in full age living in Ningbo City, Zhejiang Province in China. Their lower body types were classified and grouped according to shapes and sizes. Consequently, representative body type reflecting shapes, sizes, and characteristics were chosen and presented with their measurement values to be used as the basic data for slacks pattern making and fitness evaluation. The following are the results of the analysis: 1. Homogeneity analysis was given to determine the correspondence relationships among index types, indirect types, and size types. Index Type 1 (H-Round), Indirect Type 2 (Adjacent Straight Legs & Slight Slant of Lateral Lower Body), and Size Type 1 (Medium-Thick) were correlated; Index Type 2 (A-Trapezoid), Indirect Type 1 (Bow Legs & Slight Slant of Lateral Lower Body), and Size Type 3 (Small-Slim) were corresponded; Index Type 3 (A-Balance), Indirect Type 3 (Balance Legs & Large Slant of Lateral Lower Body), and Size Type 2 (Tall-Normal) were mutually related. 2. Mahalanobis distance based on each shape factor (index or indirect) and size factor (absolute) was obtained, while an index-absolute representative body type group (47 people, 24.4%) and an indirect-absolute representative body type group (45 people, 23.7%) were chosen, each reflecting shapes, sizes, and characteristics. 3. The representative body type group of lower body shapes of Chinese males in their 20s had average measurement values, such as 169.15cm (stature), 73.32cm (waist circumference), 75.76cm (omphalion waist circumference), 91.08cm (hip circumference), 106.02cm (outside leg length), 53.96cm (thigh circumference), and 74.42cm (crotch length).

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.35 no.12B
• /
• pp.1766-1774
• /
• 2010

This paper proposes a real-time algorithm of recognizing speed limit signs for intelligent vehicles. Contrary to previous works which use all the pixel values in the ROI (Region Of Interest) after preprocessing image at ROI and need a lot of operations, the proposed algorithm uses fewer DCT coefficients in the ROI as features of each image to reduce the number of operations. Choosing a portion of DCT coefficients which satisfy discriminant criteria for recognition, the proposed algorithm recognizes the speed limit signs using the information obtained in the selected features through LDA and MD. It selects one having the highest probability among the recognition results calculated by accumulating the classification results of consecutive individual frames. Experimental results show that the recognition rate for consecutive frames reaches to 100% with test images. When compared with the previous algorithm, the numbers of multiply and add operations are reduced by 58.6% and 38.3%, respectively.

• Safety and Health at Work
• /
• v.3 no.4
• /
• pp.278-286
• /
• 2012

Objectives: Authors investigated the pattern of the rate of occupational injuries and illnesses (ROII) at the level of enterprises in order to build a network for exchange of experience and knowledge, which would contribute to workers' safety and health through safety climate of workplace. Methods: Occupational accidents were analyzed at the manufacturing work site unit. A two step clustering process for the past patterns regarding the ROII from 2001 to 2009 was investigated. The ROII patterns were categorized based on regression analysis and the patterns were further divided according to the subtle changes with Mahalanobis distance and Ward's linkage. Results: The first clustering of ROII through regression analysis showed 5 different functions; 29 work sites of the linear function, 50 sites of the quadratic function, 95 sites of the logarithm function, 62 sites of the exponential function, and 54 sites of the sine function. Fourteen clusters were created in the second clustering. There were 3 clusters in each function categorized in the first clustering except for sine function. Each cluster consisted of the work sites with similar ROII patterns, which had unique characteristics. Conclusion: The five different patterns of ROII suggest that tailored management activities should be applied to every work site. Based on these differences, the authors selected exemplary work sites and built a network to help the work sites to share information on safety climate and accident prevention measures. The causes of different patterns of ROII, building network and evaluation of this management model should be evaluated as future researches.