• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.205-225
• /
• 2018

Convolutional Neural Network (ConvNet) is one class of the powerful Deep Neural Network that can analyze and learn hierarchies of visual features. Originally, first neural network (Neocognitron) was introduced in the 80s. At that time, the neural network was not broadly used in both industry and academic field by cause of large-scale dataset shortage and low computational power. However, after a few decades later in 2012, Krizhevsky made a breakthrough on ILSVRC-12 visual recognition competition using Convolutional Neural Network. That breakthrough revived people interest in the neural network. The success of Convolutional Neural Network is achieved with two main factors. First of them is the emergence of advanced hardware (GPUs) for sufficient parallel computation. Second is the availability of large-scale datasets such as ImageNet (ILSVRC) dataset for training. Unfortunately, many new domains are bottlenecked by these factors. For most domains, it is difficult and requires lots of effort to gather large-scale dataset to train a ConvNet. Moreover, even if we have a large-scale dataset, training ConvNet from scratch is required expensive resource and time-consuming. These two obstacles can be solved by using transfer learning. Transfer learning is a method for transferring the knowledge from a source domain to new domain. There are two major Transfer learning cases. First one is ConvNet as fixed feature extractor, and the second one is Fine-tune the ConvNet on a new dataset. In the first case, using pre-trained ConvNet (such as on ImageNet) to compute feed-forward activations of the image into the ConvNet and extract activation features from specific layers. In the second case, replacing and retraining the ConvNet classifier on the new dataset, then fine-tune the weights of the pre-trained network with the backpropagation. In this paper, we focus on using multiple ConvNet layers as a fixed feature extractor only. However, applying features with high dimensional complexity that is directly extracted from multiple ConvNet layers is still a challenging problem. We observe that features extracted from multiple ConvNet layers address the different characteristics of the image which means better representation could be obtained by finding the optimal combination of multiple ConvNet layers. Based on that observation, we propose to employ multiple ConvNet layer representations for transfer learning instead of a single ConvNet layer representation. Overall, our primary pipeline has three steps. Firstly, images from target task are given as input to ConvNet, then that image will be feed-forwarded into pre-trained AlexNet, and the activation features from three fully connected convolutional layers are extracted. Secondly, activation features of three ConvNet layers are concatenated to obtain multiple ConvNet layers representation because it will gain more information about an image. When three fully connected layer features concatenated, the occurring image representation would have 9192 (4096+4096+1000) dimension features. However, features extracted from multiple ConvNet layers are redundant and noisy since they are extracted from the same ConvNet. Thus, a third step, we will use Principal Component Analysis (PCA) to select salient features before the training phase. When salient features are obtained, the classifier can classify image more accurately, and the performance of transfer learning can be improved. To evaluate proposed method, experiments are conducted in three standard datasets (Caltech-256, VOC07, and SUN397) to compare multiple ConvNet layer representations against single ConvNet layer representation by using PCA for feature selection and dimension reduction. Our experiments demonstrated the importance of feature selection for multiple ConvNet layer representation. Moreover, our proposed approach achieved 75.6% accuracy compared to 73.9% accuracy achieved by FC7 layer on the Caltech-256 dataset, 73.1% accuracy compared to 69.2% accuracy achieved by FC8 layer on the VOC07 dataset, 52.2% accuracy compared to 48.7% accuracy achieved by FC7 layer on the SUN397 dataset. We also showed that our proposed approach achieved superior performance, 2.8%, 2.1% and 3.1% accuracy improvement on Caltech-256, VOC07, and SUN397 dataset respectively compare to existing work.

• Journal of Korean Society of Forest Science
• /
• v.9 no.1
• /
• pp.1-44
• /
• 1969

The important results which have been obtained in the investigation can be recapitulated as follows. 1. As demostrated by the experimental results and analyses concerning their effects in the on-ground type mushroom house, the constructions in relation to the side wall and ceiling of the experimental houses showed a sufficient heat insulation on effect to protect insides of the houses from outside climatic conditions. 2. As the effect on the solar type experimental mushroom house which was constructed in a half basement has been shown by the experimental results and analyses, it has been proved to be effective for making use of solar heat. However there were found two problems to be improved for putting solar houses to practical use in the farm mushroom growing: (1) the construction of the roof and ceiling should be the same as for the on-ground type house, and (2) the solar heat generating system should be reconstructed properly. A trial solar heat generating system is shown in Fig. 40. 3. Among several ventilation systems which have been studied in the experiments, the underground earthen pipe and ceiling ventilation, and vertical side wall and ceiling ventilation systems have been proved to be most effective for natural ventilation. 4. The experimental results have shown that ventilation systems such as the vertical side wall and underground ventilation systems are suitable to put to practical use as natural ventilation systems for farm mushroom houses. These ventilation systems can remarkably improve the temperature of fresh air which is introduced into the house by heat transfers within the ventilation passages, so as to approach to the desired temperature of the house without any cooling or heating operation. For example, if it is assuming that x is the outside temperature and y is the amount of temperature adjustment made by the influence of the ventilation system, the relationships that exist between x and y can be expressed by the following regression lines. Underground iron pipe ventilation system ${\cdots}{\cdots}$ y=0.9x-12.8 Underground earthen pipe ventilation system ${\cdots}{\cdots}$y=0.96x-15.11 Vertical side wall ventilation system${\cdots}{\cdots}$ y=0.94x-17.57 5. The experimental results have shown that the relationships existing between the admitted and expelled air and the $Co_2$ concentration can be described with experimental regression lines or an exponent equation as follows: 1) If it is assumed that x is an air speed cm/sec. and y is an expelled air speed in cm/sec. in a natural ventilation system, since the y is a function of the x, the relationships that exist between x and y can be expressed by the regression lines shown below: 2) If it is assumed that x is an admitted volume of air in $m^3/hr$ and y is an expelled volume of air in $m^3/hr$ in a natural ventilation system, since the y is a function of the x, the relationships that exist between x and y can be expressed by the regression lines shown below. 3) If it is assumed that the expelled air speed in cm/sec and replacement air speed in cm/sec. at the bed surface in a natural ventilation system are shown as x and y, respectively, since the y is a function of the x, the relationships that exist between x and y can be expressed by the following regression line: G.E. (100%)- C.V. (50%) ventilation system${\cdots}$ y=0.54X+0.84 4) If it is assumed that the replacement air speed in cm/sec. at the bed surface is shown as x, and $CO_2$ concentration which is expressed by multiplying 1000 times the actual value of $CO_2$ % is shown as y, in a natural ventilation system, since the y is a function of the x the relationships that exist between x and y can be expressed by the following regression line: G.E. (100%)- C.V. (50%) ventilation system${\cdots}{\cdots}$ y=114.53-6.42x 5) If it is assumed that the expelled volume of air is shown as x and the $CO_2$ concentration which is expressed by multiplying 1000 times the actual of $CO_2$ % is shown as y in a natural ventilation system, since the y is a function of of the x, the relationships that exist between x and y can be expressed by the following exponent equation: G.E. (100%)-C.V. (50%) ventilation system${\cdots}{\cdots}$ $$y=127.18{\times}1.0093^{-X}$$ 6. The experimental results have shown that the ratios of the crass sectional area of the G.E. and C.V. vent to the total cubic capacity of the house, required for providing an adequate amount of air in a natural ventilation system, can be estimated as follows: G.E. (admitting vent of the underground ventilation)${\cdots}{\cdots}$ 0.30-0.5% (controllable) C.V. (expelling vent of the ceiling ventilation)${\cdots}{\cdots}$ 0.8-1.0% (controllable) 7. Among several heating devices which were studied in the experiments, the hot-water boilor which was modified to be fitted both as hot-water toiler and as a pressureless steam-water was found most suitable for farm mushroom growing.

• Korean Journal of Environmental Agriculture
• /
• v.14 no.2
• /
• pp.232-245
• /
• 1995

The garbage from the dwelling house was composted in two kinds of small composter in the laboratory, and the possibility of garbage composting was examined. The composters were general small. One (type 3) was constructed with the double layer walls and the other (type 4) was the same as the first except for being insulated. Because it was found that type 3 was not available for composting under our meteorological conditions through the winter experiment, only type 4 was tested in spring and summer. The experiment was performed for 8 weeks in each season. The seasonal variation of several components in the compost was evaluated and discussed. The results summarized below were those obtained at the end of the experiment, if the time was not specified. 1) The maximum temperature was $43^{\circ}C$ in winter, $55^{\circ}C$ in spring and $56^{\circ}C$ in summer. 2) The mass was reduced to an average of 63% and the volume reduction was an average of 78%. 3) The density was estimated as 1.5 kg/l in winter and 0.8 kg/l in spring and summer. 4) The water content was not much changed during the composting periods. It was 79.3% in winter, 75.0% in spring and 70.0% in summer. 5) After pH value increased during the first week, it decreased until the second week and increased again continuously thereafter. It reached pH 6.19 in winter, pH 7.59 in spring and pH 8.69 in summer. 6) The faster the organic matter was decomposed, the greater the ash content increased. The contents of cellulose and lignin increased, but that of hemicellulose decreased during the composting period. 7) Nitrogen contents were in the range of 3.3-6.8% and especially high in summer. After ammonium contents increased at the early stage of the composting period, they decreased. The maximum ammonium-nitrogen content was 2,404mg/kg after 8 weeks in winter, 12,400mg/kg after 3 weeks in spring and 20,718mg/kg after 3 weeks in summer. C/N-ratios decreased with the lapse of composting time, but they were not much changed. Nitrification occurred actively in summer. 8) The contents of volatile and higher fatty acids increased at the early stage of composting and reduced after that. The maximum content of total fatty acid was 9.7% after 6 weeks in winter, 14.8% after 6 weeks in spring and 15.8% after 2 weeks in summer. 9) The contents of inorganic components were not accumulated as composting proceeded. They were in the range of 0.9-4.4% $P_2O_5$, 1.6-2.4% $K_2O$, 2.2-5.4% CaO and 0.30-0.61% MgO. 10) CN and heavy metal contents did not show any tendency. They were in the range of 0.21-14.55mg/kg CN, 11-166mg/kg Zn, 5-65mg/kg Cu, 0.5-10.8mg/kg Cd, 6- 35mg/kg Pb, ND-33 mg/kg Cr and ND-302.04 g/kg Hg.

• Journal of Global Scholars of Marketing Science
• /
• v.17 no.2
• /
• pp.31-54
• /
• 2007

The exchange between buyers and sellers in the industrial market is changing from short-term to long-term relationships. Long-term relationships are governed mainly by formal contracts or informal agreements, but many scholars are now asserting that controlling relationship by using formal contracts under environmental dynamism is inappropriate. In this case, partners will depend on each other's flexibility or interdependence. The former, flexibility, provides a general frame of reference, order, and standards against which to guide and assess appropriate behavior in dynamic and ambiguous situations, thus motivating the value-oriented performance goals shared between partners. It is based on social sacrifices, which can potentially minimize any opportunistic behaviors. The later, interdependence, means that each firm possesses a high level of dependence in an dynamic channel relationship. When interdependence is high in magnitude and symmetric, each firm enjoys a high level of power and the bonds between the firms should be reasonably strong. Strong shared power is likely to promote commitment because of the common interests, attention, and support found in such channel relationships. This study deals with environmental dynamism in high-tech industry. Firms in the high-tech industry regard it as a key success factor to successfully cope with environmental changes. However, due to the lack of studies dealing with environmental dynamism and supply chain commitment in the high-tech industry, it is very difficult to find effective strategies to cope with them. This paper presents the results of an empirical study on the relationship between environmental dynamism and supply chain commitment in the high-tech industry. We examined the effects of consumer, competitor, and technological dynamism on supply chain commitment. Additionally, we examined the moderating effects of flexibility and dependence of supply chains. This study was confined to the type of high-tech industry which has the characteristics of rapid technology change and short product lifecycle. Flexibility among the firms of this industry, having the characteristic of hard and fast growth, is more important here than among any other industry. Thus, a variety of environmental dynamism can affect a supply chain relationship. The industries targeted industries were electronic parts, metal product, computer, electric machine, automobile, and medical precision manufacturing industries. Data was collected as follows. During the survey, the researchers managed to obtain the list of parts suppliers of 2 companies, N and L, with an international competitiveness in the mobile phone manufacturing industry; and of the suppliers in a business relationship with S company, a semiconductor manufacturing company. They were asked to respond to the survey via telephone and e-mail. During the two month period of February-April 2006, we were able to collect data from 44 companies. The respondents were restricted to direct dealing authorities and subcontractor company (the supplier) staff with at least three months of dealing experience with a manufacture (an industrial material buyer). The measurement validation procedures included scale reliability; discriminant and convergent validity were used to validate measures. Also, the reliability measurements traditionally employed, such as the Cronbach's alpha, were used. All the reliabilities were greater than.70. A series of exploratory factor analyses was conducted. We conducted confirmatory factor analyses to assess the validity of our measurements. A series of chi-square difference tests were conducted so that the discriminant validity could be ensured. For each pair, we estimated two models-an unconstrained model and a constrained model-and compared the two model fits. All these tests supported discriminant validity. Also, all items loaded significantly on their respective constructs, providing support for convergent validity. We then examined composite reliability and average variance extracted (AVE). The composite reliability of each construct was greater than.70. The AVE of each construct was greater than.50. According to the multiple regression analysis, customer dynamism had a negative effect and competitor dynamism had a positive effect on a supplier's commitment. In addition, flexibility and dependence had significant moderating effects on customer and competitor dynamism. On the other hand, all hypotheses about technological dynamism had no significant effects on commitment. In other words, technological dynamism had no direct effect on supplier's commitment and was not moderated by the flexibility and dependence of the supply chain. This study makes its contribution in the point of view that this is a rare study on environmental dynamism and supply chain commitment in the field of high-tech industry. Especially, this study verified the effects of three sectors of environmental dynamism on supplier's commitment. Also, it empirically tested how the effects were moderated by flexibility and dependence. The results showed that flexibility and interdependence had a role to strengthen supplier's commitment under environmental dynamism in high-tech industry. Thus relationship managers in high-tech industry should make supply chain relationship flexible and interdependent. The limitations of the study are as follows; First, about the research setting, the study was conducted with high-tech industry, in which the direction of the change in the power balance of supply chain dyads is usually determined by manufacturers. So we have a difficulty with generalization. We need to control the power structure between partners in a future study. Secondly, about flexibility, we treated it throughout the paper as positive, but it can also be negative, i.e. violating an agreement or moving, but in the wrong direction, etc. Therefore we need to investigate the multi-dimensionality of flexibility in future research.