• Journal of Internet Computing and Services
• /
• v.20 no.4
• /
• pp.91-102
• /
• 2019

In autonomous driving systems, the ability to classify pedestrians in images captured by cameras is very important for pedestrian safety. In the past, after extracting features of pedestrians with HOG(Histogram of Oriented Gradients) or SIFT(Scale-Invariant Feature Transform), people classified them using SVM(Support Vector Machine). However, extracting pedestrian characteristics in such a handcrafted manner has many limitations. Therefore, this paper proposes a method to classify pedestrians reliably and effectively using CNN's(Convolutional Neural Network) deep features and transfer learning. We have experimented with both the fixed feature extractor and the fine-tuning methods, which are two representative transfer learning techniques. Particularly, in the fine-tuning method, we have added a new scheme, called M-Fine(Modified Fine-tuning), which divideslayers into transferred parts and non-transferred parts in three different sizes, and adjusts weights only for layers belonging to non-transferred parts. Experiments on INRIA Person data set with five CNN models(VGGNet, DenseNet, Inception V3, Xception, and MobileNet) showed that CNN's deep features perform better than handcrafted features such as HOG and SIFT, and that the accuracy of Xception (threshold = 0.5) isthe highest at 99.61%. MobileNet, which achieved similar performance to Xception and learned 80% fewer parameters, was the best in terms of efficiency. Among the three transfer learning schemes tested above, the performance of the fine-tuning method was the best. The performance of the M-Fine method was comparable to or slightly lower than that of the fine-tuningmethod, but higher than that of the fixed feature extractor method.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.1
• /
• pp.36-43
• /
• 2021

In this research, the accuracy of YOLO v3 algorithm in object detection during AGV (Automated Guided Vehicle) operation was investigated. First of all, AGV with 2D LiDAR and stereo camera was prepared. AGV was driven along the route scanned with SLAM (Simultaneous Localization and Mapping) using 2D LiDAR while front objects were detected through stereo camera. In order to evaluate the accuracy of YOLO v3 algorithm, recall, AP (Average Precision), and mAP (mean Average Precision) of the algorithm were measured with a degree of machine learning. Experimental results show that mAP, precision, and recall are improved by 10%, 6.8%, and 16.4%, respectively, when YOLO v3 is fitted with 4000 training dataset and 500 testing dataset which were collected through online search and is trained additionally with 1200 dataset collected from the stereo camera on AGV.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.34 no.4
• /
• pp.428-433
• /
• 1989

The objective of this experiment was to determine if there were feasible to transplant infant rice seedling for machine transplanting. Cultivars tested were Sobaegbyeo and Daecheongbyeo, ]aponica type cultivars. Infant seedling, young seedling and semi-adult seedling were raised with sailor rock wool in seedling tray from 1986 to 1988. Infant rice seedling raised more than 4 days after sowing with rock wool was uniform, and low in ratio of missing hill at machine transplanting. Tiller number per m$^2$ was more in infant rice seedling, young seedling and semi-adult seedling, in that order. Heading dates were not significantly different among seedlings applied, however rice plant sown directly was later 4 or 5 days than any other seedlings. Panicle number per m$^2$ was more in infant seedling than in semi-adult seedling, but ripened ratio was lower in infant seedling due to lodging. Thus there were not greatly different in yield among seedlings tested. Therefore infant rice seedling (more than 5cm in plumule length) raised for 7 days was most optimum, and rock wool would be used as a nursery soil instead of nursery soil for raising infant rice seedling in machine transplanting.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.39 no.5
• /
• pp.495-501
• /
• 1994

This experiment was carried out to investigate the proper seeding methed for direct seeding culture of rice in puddled soil at ill-drained paddy field. Mangeumbyeo was seeded on May 11 with drill seeder attached to cultivater, dibbling seeder attached to transplanter and power dust blower. And to determine the proper flooding time after seeding, Mangeumbyeo was seeded on May 7 and June 1 with drill seeder. Emergence ratio wasn't significantly different between drill seeder seeding and dibbling seeder seeding but emergence was very low with power dust blower. Working hour for seeding was shortened in the order of power dust blower, drill seeder and dibbling seeder, and these seeding methods saved the working hour remarkably compared with machine transplanting of infant seedling by omitting the procedures of raising seedling and transplanting. Lodging occurred slightly in all seeding methods. However, yield wasn't significantly different between machine transplanting of infant seeding and direct seeding culture in puddled soil except power dust blower. When field was flooded on seeding date, the emergence period was shortened, emergence ratio was increased and weed occurrence was decreased. Yield was highest when flooding was done on seeding date as the number of panicle per m$^2$ and the number of spikelets per m$^2$ were higher than any other flooding time.

• Journal of the Korean Society of Manufacturing Technology Engineers
• /
• v.22 no.2
• /
• pp.318-323
• /
• 2013

For fast and accurate motion of 6-axis articulated robot, more noble motion control strategy is needed. In general, the movement strategy of industrial robots can be divided into two kinds, PTP (Point to Point) and CP (Continuous Path). Recently, industrial robots which should be co-worked with machine tools are increasingly needed for performing various jobs, as well as simple handling or welding. Therefore, in order to cope with high-speed handling of the cooperation of industrial robots with machine tools or other devices, CP should be implemented so as to reduce vibration and noise, as well as decreasing operation time. This paper will realize CP motion (especially joint-linear) blending in 3-dimensional space for a 6-axis articulated (lab-manufactured) robot (called as "RS2") by using LabVIEW$^{(R)}$ (6) programming, based on a parametric interpolation. Another small contribution of this paper is the proposal of motion blending simulation technique based on Recurdyn$^{(R)}$ V7 and Solidworks$^{(R)}$, in order to figure out whether the joint-linear blending motion can generate the stable motion of robot in the sense of velocity magnitude at the end-effector of robot or not. In order to evaluate the performance of joint-linear motion blending, simple PTP (i.e., linear-linear) is also physically implemented on RS2. The implementation results of joint-linear motion blending and PTP are compared in terms of vibration magnitude and travel time by using the vibration testing equipment of Medallion of Zonic$^{(R)}$. It can be confirmed verified that the vibration peak of joint-linear motion blending has been reduced to 1/10, compared to that of PTP.

• Journal of Korea Foundry Society
• /
• v.26 no.4
• /
• pp.174-179
• /
• 2006

This study aims to investigate the microstructures and mechanical properties of DCI manufactured by sand and metal mold casting. To prohibit the formation of the chill, carbon, silicon and rare earth($0{\sim}0.2\;wt%$) were controlled and temperature of metal mold was constantly kept at $160^{\circ}C$. The sizes, counts and nodularity ratios of nodules were analyzed by image analysis device. Wear test using pin-on-disc wear tester was carried out under the conditions of load 47.2N, velocity 0.4 m/s and distance 2000 m. Tensile test using Instron type testing machine was performed with velocity of 0.1 mm/min according to the KS B 0802. The formation of the chill was not observed when percentage of the carbon and silicon were 3.8 and 2.5. Mechanical properties of GCD manufactured by metal mold were better than sand casting.

• Proceedings of the KIEE Conference
• /
• 1999.11d
• /
• pp.912-914
• /
• 1999

Composite insulation system of liquid nitrogen and solid spacer is widely applied in high temperature superconducting power machine. This study has three step procedure. As follow, first step is composition of parallel deposited electrode and vertically deposited electrode along the direction of immersion in liquid nitrogen($LN_2$). Second step is investigation into surface flashover voltage of solid spacer under partially immersed in $LN_2$, and last step is comparison the result of this research with that of fully immersed in $LN_2$ and at cryogenic temperature gaseous nitrogen($GN_2$). This result presented that surface flashover voltage along solid spacer half immersed in $LN_2$ was almost the same as that of fully immersed spacer when the thickness of spacer(t) was t<10 mm. In the case of t>10 mm, however, spacer flashover voltage was equal to that obtained in $GN_2$ at cryogenic temperature.

• Journal of Bio-Environment Control
• /
• v.6 no.3
• /
• pp.183-189
• /
• 1997

This study was performed to Investigate size of droplet sprinkled from mini-sprinkler. Twelve different kinds of the sprinkler having various structures and sizes of nozzle orifices were selected and tested. Diameters of the droplet reached at several distances from a sprinkler were measured by a machine vision system and the volume median diameters (VMM) were determined statistically. The size of droplet was not affected much by the size of nozzle orifice of a sprinkler but was rather more affected by structure of the sprinkler, especially by the shape of spreader of the sprinkler. Experiment of varying pressure of sprinkling water validated that the size of droplet was inversely proportional to water pressure powered by 1/3. Hence the size of droplet at any water pressure could be easily estimated from experimental data. The size of droplet increased as travel distance of the droplet increases in a relationship of and order function. The size of droplet of the tested sprinkler were in the ranges of 100-300fm within 1m of droplet travel distance, 230~470${\mu}{\textrm}{m}$ within 1~2m of droplet travel distance and 300~770${\mu}{\textrm}{m}$ within 2~3m of droplet travel distance.

• Korean Journal of Weed Science
• /
• v.18 no.1
• /
• pp.12-19
• /
• 1998

The pattern of rice cultivation in Korea is changing from hand transplanting and machine transplanting into direct-seeding for low input and cost-down practices. However, there are still some obstacles to establish the direct-seeding cultural practices because of poor seedling stand, lodging, and weed infestation. In particular, there were much more and wide weed occurrence in direct-seeding rice as compared with the transplanted rice. Weed occurrence in terms of dry weight of weeds was highest in dry direct-seeded rice followed by sowing an puddled soil, sowing on submerged soil in order as compared with transplanting with infant rice seedling. Echinochloa crux-galli was a common dominant weed with 35~44% distribution in any direct-seeding methods. The dominant weed species were E. crux-galli, Cyperus difformis, C. serotinus, and Ludwigia prostrata in dry-seeded rice. The dominant weed species in water-seeded rice were E. crus-galli, Eleocharis kuroguwai, Scirpus juncoides, and Monochoria vaginalis. Yield loss of rice due to weeds at weedy condition was 96% in dry direct-seeded rice. In water-seeded rice, yield loss was 61%, whereas yield loss was 40% in machine transplanting with infant rice seedling.

• Journal of Surface Science and Engineering
• /
• v.34 no.1
• /
• pp.62-67
• /
• 2001

Green technology is being developed up to a point that is feasible not only in an environmental sense, but also in an economical viewpoint. This paper introduces two case studies that applied $CO_2$ technology into nuclear industry. 1) Nuclear laundry : A laundry machine that uses liquid and supercritical $CO_2$ as a solvent for decontamination of contaminated working dresses in nuclear power plants was developed. The machine consists of a 16 liter reactor, a recovery system with compressors, and storage tanks. All $CO_2$ used in cleaning is fully recovered and reused in next cleaning, resulting in no production of secondary nuclear waste. Decontamination factor is still lower than that in the methods currently used in the plant. Nuclear laundry using $CO_2$ looks promising with technical improvements-surfactants and mechanical agitation. 2) $CO_2$ nozzle decontamination : An adjustable nozzle for controlling the size of dry ice snow was developed. Using the developed nozzle, a surface decontamination device was made. Human oils like fingerprints on glass were easy to remove. Decontamination ability was tested using a contaminated pump-housing surface. About 40 to 80% of radioactivity was removed. This device is effective in surface-decontamination of any electrical devices like detector, controllers which cannot be cleaned in aqueous solution.