• 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.

• The Korean Journal of Air & Space Law and Policy
• /
• v.18
• /
• pp.135-183
• /
• 2003

Aerospace force is a determining factor in a modem war. The combat field is expanding to space. Thus, the legitimacy of establishing aerospace force is no longer an debating issue, but "how should we establish aerospace force" has become an issue to the military. The standard limiting on the military use of space should be non-aggressive use as asserted by the U.S., rather than non-military use as asserted by the former Soviet Union. The former Soviet Union's argument is not even strongly supported by the current Russia government, and realistically is hard to be applied. Thus, the multi-purpose satellite used for military surveillance or a commercial satellite employed for military communication are allowed under the U.S. principle of peaceful use of space. In this regard, Air Force may be free to develop a military surveillance satellite and a communication satellite with civilian research institute. Although MTCR, entered into with the U.S., restricts the development of space-launching vehicle for the export purpose, the development of space-launching vehicle by the Korea Air Force or Korea Aerospace Research Institute is beyond the scope of application of MTCR, and Air Force may just operate a satellite in the orbit for the military purpose. The primary task for multi-purpose satellite is a remote sensing; SAR sensor with high resolution is mainly employed for military use. Therefore, a system that enables Air Force, the Korea Aerospace Research Institute, and Agency for Defense Development to conduct joint-research and development should be instituted. U.S. Air Force has dismantled its own space-launching vehicle step by step, and, instead, has increased using private space launching vehicle. In addition, Military communication has been operated separately from civil communication services or broadcasting services due to the special circumstances unique to the military setting. However, joint-operation of communication facility by the military and civil users is preferred because this reduces financial burden resulting from separate operation of military satellite. During the Gulf War, U.S. armed forces employed commercial satellites for its military communication. Korea's participation in space technology research is a little bit behind in time, considering its economic scale. In terms of budget, Korea is to spend 5 trillion won for 15 years for the space activities. However, Japan has 2 trillion won annul budget for the same activities. Because the development of space industry during initial fostering period does not apply to profit-making business, government supports are inevitable. All space development programs of other foreign countries are entirely supported by each government, and, only recently, private industry started participating in limited area such as a communication satellite and broadcasting satellite, Particularly, Korea's space industry is in an infant stage, which largely demands government supports. Government support should be in the form of investment or financial contribution, rather than in the form of loan or borrowing. Compared to other advanced countries in space industry, Korea needs more budget and professional research staff. Naturally, for the efficient and systemic space development and for the prevention of overlapping and distraction of power, it is necessary to enact space-related statutes, which would provide dear vision for the Korea space development. Furthermore, the fact that a variety of departments are running their own space development program requires a centralized and single space-industry development system. Prior to discussing how to coordinate or integrate space programs between Agency for Defense Development and the Korea Aerospace Research Institute, it is a prerequisite to establish, namely, "Space Operations Center"in the Air Force, which would determine policy and strategy in operating space forces. For the establishment of "Space Operations Center," policy determinations by the Ministry of National Defense and the Joint Chief of Staff are required. Especially, space surveillance system through using a military surveillance satellite and communication satellite, which would lay foundation for independent defense, shall be established with reference to Japan's space force plan. In order to resolve issues related to MTCR, Air Force would use space-launching vehicle of the Korea Aerospace Research Institute. Moreover, defense budge should be appropriated for using multi-purpose satellite and communication satellite. The Ministry of National Defense needs to appropriate 2.5 trillion won budget for space operations, which amounts to Japan's surveillance satellite operating budges.

• Tuberculosis and Respiratory Diseases
• /
• v.42 no.6
• /
• pp.846-854
• /
• 1995

Background: Cytokeratin 19 is a subunit of cytokeratin intermediate filament expressed in simple epithelia such as respiratory epithelial cells and their malignant counterparts. An immunoradiometric assay is available to detect a fragment of the cytokeratin, referred to as Cyfra 21-1 in the serum. This study was conducted to evaluate the clinical utility of this new marker in the diagnosis of lung cancer compared with established markers of squamous cell carcinoma antigen (SCC Ag) and carcino-embryonic antigen(CEA). In addition, we compared the diagnostic sensitivity and specificity of Cyfra 21-1 with those of SCC Ag in squamous cell carcinoma of the lung. We also measured the level of Cyfra 21-1 in the different stages of squamous cell carcinoma of the lung. Method: We measured Cyfra 21-1(ELSA-CYFRA 21-1), SCC Ag(ABBOTT SCC RIABEAD) and CEA(ELSA2-CEA) in 79 patients with primary lung cancer and in 78 persons as a comparison group including 32 patients with pulmonary tuberculosis, 23 patients with benign lung disease and 23 cases with healthy individual. Cyfra 21-1 is measured by a solid-phase immunoradiometric assay(CIS Bio International, France) based on the two-site sandwich method. SCC Ag is measured by a radioimmunoassay(Abbott Laboratories, USA). CEA is measured by a immunoradiometric assay(CIS Bio International, France). All data were expressed as the mean$\pm$standard deviation. Results: 1) The mean value of Cyfra 21-1 was $18.38{\pm}3.65\;ng/mL$ in the lung cancer and $1.l6{\pm}0.53\;ng/mL$ in the comparison group(p<0.0001). SCC Ag was $3.53{\pm}6.06\;ng/mL$ in the lung cancer and $1.19{\pm}0.5\;ng/mL$ in the comparison group(p<0.01). CEA was $35.03{\pm}13.9\;ng/mL$ in the lung cancer and $2.89{\pm}1.01\;ng/mL$ in the comparison group(p<0.0001). 2) Cyfra 21-1 level in squamous cell carcinoma($31.52{\pm}40.13\;ng/mL$) was higher than that in adenocarcinoma($2.41{\pm}1.34\;ng/mL$)(p<0.0001) and small cell carcinoma($2.15{\pm}2.05\;ng/mL$)(p=0.007). SCC Ag level in squamous cell carcinoma($5.1{\pm}7.68\;ng/mL$) was higher than that in adenocarcinoma($1.36{\pm}0.69\;ng/mL$)(p=0.009) and small cell carcinoma($1.1{\pm}0.24\;ng/mL$) (p=0.024). 3) The level of Cyfra 21-1 was not correlated with the progression of stage in squamous cell carcinoma of the lung. 4) Using the cut-off value of 3.3ng/mL, the diagnostic sensitivity of Cyfra 21-1 was 83% in squamous cell carcinoma, 22% in adenocarcinoma and 17% in small cell carcinoma. The sensitivity of SCC Ag and CEA were 39% and 20%, respectively in squamous cell carcinoma, 11% and 39% in adenocarcinoma, and 0% and 33% in small cell carcinoma. 5) Comparison of the receiver operating characteristics curves(ROC curve) for Cyfra 21-1, SCC Ag and CEA revealed that Cyfra 21-1 showed highest diagnostic sensitivity among them in the diagnosis of lung cancer. Conclusion: Cyfra 21-1 is thought to be a better tumor marker for the diagnosis of lung cancer than SCC Ag and CEA, especially in squamous cell carcinoma of the lung.