• Journal of Korean Society of Forest Science
• /
• v.109 no.1
• /
• pp.23-30
• /
• 2020

The use of increased computing power, machine learning, and deep learning techniques have dramatically increased in various sectors. In particular, image detection algorithms are broadly used in forestry and remote sensing areas to identify forest types and tree species. However, in South Korea, machine learning has rarely, if ever, been applied in forestry image detection, especially to classify tree species. This study integrates the application of machine learning and forest image detection; specifically, we compared the ability of two machine learning data collection methods, namely image data captured by forest experts (D1) and web-crawling (D2), to automate the classification of five trees species. In addition, two methods of characterization to train/test the system were investigated. The results indicated a significant difference in classification accuracy between D1 and D2: the classification accuracy of D1 was higher than that of D2. In order to increase the classification accuracy of D2, additional data filtering techniques were required to reduce the noise of uncensored image data.

• Korean Journal of Remote Sensing
• /
• v.24 no.6
• /
• pp.621-629
• /
• 2008

This study was carried out in part of Tamil Nadu, India. Also, Landsat ETM+ image and field sampling data were acquired. The field data were basal area, number of trees and number of species. Using the data set, this study performed a three steps processing, (1) Image classification (2) extracting the vegetation indices(NDVI, Tasseled cap brightness, greenness and wetness) (3) mapping the prediction of biodiversity distribution using basal area and NDVI image value. Basal area was significantly correlated with NDVI. The result of classification showed 69% overall accuracy.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.57 no.4
• /
• pp.1-9
• /
• 2015

The development of drought index that provides detailed-spatial-resolution drought information is essential for improving drought planning and preparedness. The objective of this study was to develop the concept of using satellite-based hybrid drought index called the Vegetation Drought Response Index in South Korea (VegDRI-SKorea) that could improve spatial resolution for monitoring local and regional drought. The VegDRI-SKorea was developed using the Classification And Regression Trees (CART) algorithm based on remote sensing data such as Normalized Difference Vegetation Index (NDVI) from MODIS satellite images, climate drought indices such as Self Calibrating Palmer Drought Severity Index (SC-PDSI) and Standardized Precipitation Index (SPI), and the biophysical data such as land cover, eco region, and soil available water capacity. A case study has been done for the 2012 drought to evaluate the VegDRI-SKorea model for South Korea. The VegDRI-SKorea represented the drought areas from the end of May and to the severe drought at the end of June. Results show that the integration of satellite imageries and various associated data allows us to get improved both spatially and temporally drought information using a data mining technique and get better understanding of drought condition. In addition, VegDRI-SKorea is expected to contribute to monitor the current drought condition for evaluating local and regional drought risk assessment and assisting drought-related decision making.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38A no.1
• /
• pp.1-9
• /
• 2013

This paper propose a human action recognition method that uses bag-of-features (BoF) based on CS-LBP (center-symmetric local binary pattern) and a spatial pyramid in addition to the random forest classifier. To construct the BoF, an image divided into dense regular grids and extract from each patch. A code word which is a visual vocabulary, is formed by k-means clustering of a random subset of patches. For enhanced action discrimination, local BoF histogram from three subdivided levels of a spatial pyramid is estimated, and a weighted BoF histogram is generated by concatenating the local histograms. For action classification, a random forest, which is an ensemble of decision trees, is built to model the distribution of each action class. The random forest combined with the weighted BoF histogram is successfully applied to Standford Action 40 including various human action images, and its classification performance is better than that of other methods. Furthermore, the proposed method allows action recognition to be performed in near real-time.

• Korean Journal of Environment and Ecology
• /
• v.26 no.3
• /
• pp.382-393
• /
• 2012

The purpose of this study was to research the vegetation structure of the Bongamsa Forest Genetic Resourses Resreve Area in Mt. Heuyang, Mungyeong, Gyeongsangbuk-do. For doing this, ninety-two plots($100m^2/plot$) were set up and investigated, and then Classification analysis and Ordination analysis were carried out. As a result, the vegetation of this area is divided to nine communities; Quercus mongolica community, Quercus variabilis community, Pinus densiflora community, Pinus densiflora-Quercus serrata community, Pinus densiflora-Quercus mongolica community, Quercus serrata community, Decideous broad leaf community, Pinus koraiensis community, Larix kaempferi community. To compare between the communities, statistical analyses were conducted with topographical condition and the results of the vegetational investigation of each community. In altitude, slope, the number of species, the number of individuals in canopy and the number of individuals of understory layer, the mean averages among the communities were different in statistically significance. Then we analysed the vegetation community structure with Importance Percentage of each stratum. The oak tree communities were expected to keep or expand the actual communities because oak trees are spread widely in canopy and understory layers. But the pine tree dominant communities were expected to be succeeded to oak tree communities in the future because of the wide expansion of oak trees.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.15 no.2
• /
• pp.113-125
• /
• 2012

This research is aimed at evaluating the land surface characterization of KOMPSAT-3A middle infrared (MIR) data. Airborne Hyperspectral Scanner (AHS) data, which has MIR bands with high spatial resolution, were used to assess land surface temperature (LST) retrieval and classification accuracy of MIR bands. Firstly, LST values for daytime and nighttime, which were calculated with AHS thermal infrared (TIR) bands, were compared to digital number of AHS MIR bands. The determination coefficient of AHS band 68 (center wavelength $4.64{\mu}m$) was over 0.74, and was higher than other MIR bands. Secondly, The land cover maps were generated by unsupervised classification methods using the AHS MIR bands. Each class of land cover maps for daytime, such as water, trees, green grass, roads, roofs, was distinguished well. But some classes of land cover maps for nighttime, such as trees versus green grass, roads versus roofs, were not separated. The image classification using the difference images between daytime AHS MIR bands and nighttime AHS MIR bands were conducted to enhance the discrimination ability of land surface for AHS MIR imagery. The classification accuracy of the land cover map for zone 1 and zone 2 was 67.5%, 64.3%, respectively. It was improved by 10% compared to land cover map of daytime AHS MIR bands and night AHS MIR bands. Consequently, new algorithm based on land surface characteristics is required for temperature retrieval of high resolution MIR imagery, and the difference images between daytime and nighttime was considered to enhance the ability of land surface characterization using high resolution MIR data.

• Korean Journal of Environment and Ecology
• /
• v.27 no.4
• /
• pp.449-461
• /
• 2013

This study is contributed to examine the vegetation characteristics of the Quercus mongolica which are located in between two summits, hyangnobong and gitdaebaggybong, with dividing two regions(I, II). A total of 63 Quercus mongolica communities were enrolled in this survey. A~E, 5 communities were the result of analysed classification by TWINSPAN. Spatial distribution of the communities; 'A' community is concentrated at an altitude of about 1,300 meters in Mt. Taebaek(II region). 'B' and 'C' communities are concentrated at an altitude of between 700 and 1500 meters in Mt. Seorak and Odae(Iregion). 'D' and 'E' communities are concentrated at an altitude of below 1,100 meters in Mt. Cheongok and geumdaebong(II region), and also Mt. Seorak and Odae(Iregion). Arboral and shrub populations, maximum species diversity and height of arboral of each communities have a characteristic of decreasing according to increasing altitude gradually. And the result of studying on arboreal growth increment with up to 40-year-old trees, there is no special tendency with growth increment in the altitude of subnormal 1,100 meters, but there is a tendency that growth increment becomes lower in the altitude of more than 1,100 meters. In correlation analysis, it is -0.53 in the Quercus mongolica and Betula costata community at an altitude of over 1,000 meters, the correlation shows negative(-). It is +0.59 in the Quercus mongolica and Rubus crataegifolius community at an altitude of less than 1,000 meters, the correlation shows positive(+). And those are cognate trees with the correlation of Betula costata, Betula costata Trautv., Tripterygium regelii Sprague&Takeda., Acer barbinerve Maxim. With a summary of the results, the changes in the environment according to the elevation affect Quercus mongolica community.

• Journal of Korean Society of Forest Science
• /
• v.112 no.2
• /
• pp.195-208
• /
• 2023

This study aimed to use three-dimensional point cloud data (PCD) obtained from Terrestrial Laser Scanning (TLS) and Mobile Laser Scanning (MLS) to evaluate a deep learning-based species classification model for two tree species: Pinus koraiensis and Larix kaempferi. Sixteen models were constructed based on the three conditions: LiDAR platform (TLS and MLS), down-sampling intensity (1024, 2048, 4096, 8192), and deep learning model (PointNet, PointNet++). According to the classification accuracy evaluation, the highest kappa coefficients were 93.7% for TLS and 96.9% for MLS when applied to PCD data from the PointNet++ model, with down-sampling intensities of 8192 and 2048, respectively. Furthermore, PointNet++ was consistently more accurate than PointNet in all scenarios sharing the same platform and down-sampling intensity. Misclassification occurred among individuals of different species with structurally similar characteristics, among individual trees that exhibited eccentric growth due to their location on slopes or around trails, and among some individual trees in which the crown was vertically divided during tree segmentation.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.5
• /
• pp.48-55
• /
• 2009

Prior research works in indoor-outdoor classification have been conducted based on a simple combination of low-level features. However, since there are many challenging problems due to the extreme variability of the scene contents, most methods proposed recently tend to combine the low-level features with high-level information such as the presence of trees and sky. To extract these regions from videos, we need to conduct additional tasks, which may yield the increasing number of feature dimensions or computational burden. Therefore, an efficient indoor-outdoor scene classification method is proposed in this paper. First, the video is divided into the five same-sized blocks. Then we define and use the edge and color orientation histogram (ECOH) descriptors to represent each sub-block efficiently. Finally, all ECOH values are simply concatenated to generated the feature vector. To justify the efficiency and robustness of the proposed method, a diverse database of over 1200 videos is evaluated. Moreover, we improve the classification performance by using different weight values determined through the learning process.

• Korean Journal of Remote Sensing
• /
• v.36 no.6_2
• /
• pp.1591-1604
• /
• 2020

This study is a study on classifying land cover by applying high-resolution satellite images to deep learning algorithms and verifying the performance of algorithms for each spatial object. For this, the Fully Convolutional Network-based algorithm was selected, and a dataset was constructed using Kompasat-3 satellite images, land cover maps, and forest maps. By applying the constructed data set to the algorithm, each optimal hyperparameter was calculated. Final classification was performed after hyperparameter optimization, and the overall accuracy of DeeplabV3+ was calculated the highest at 81.7%. However, when looking at the accuracy of each category, SegNet showed the best performance in roads and buildings, and U-Net showed the highest accuracy in hardwood trees and discussion items. In the case of Deeplab V3+, it performed better than the other two models in fields, facility cultivation, and grassland. Through the results, the limitations of applying one algorithm for land cover classification were confirmed, and if an appropriate algorithm for each spatial object is applied in the future, it is expected that high quality land cover classification results can be produced.