• Korean Journal of Optics and Photonics
• /
• v.22 no.4
• /
• pp.184-190
• /
• 2011

A compact imaging spectrometer (COMIS) was developed for a microsatellite STSAT-3. The satellite is now rescheduled to be launched into a low sun-synchronous Earth orbit (~700 km) by the end of 2012. Its main operational goal is the imaging of the Earth's surface and atmosphere with ground sampling distance of 27 m and 2 - 15 nm spectral resolution over visible and near infrared spectrum (0.4 - 1.05 ${\mu}m$). A flight model of COMIS was developed following an engineering model that had successfully demonstrated hyperspectral imaging capability and structural rigidity. In this paper we report the environmental test results of the flight model. The mechanical stiffness of the model was confirmed by a small shift of the natural frequency i.e., < 1% over 10 gRMS random vibration test. Electrical functions of the model were also tested without showing any anomalies during and after vacuum thermal cycling test with < $10^{-5}$ torr and $-30^{\circ}C\;-\;35^{\circ}C$. The imaging capability of the model, represented by a modulation transfer function (MTF) value at the Nyquist frequency, was also kept unvaried after all those environmental tests.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.12 no.12
• /
• pp.2349-2356
• /
• 2008

In this paper, we present the method to effectively extract and classify the EEG caused by only brain activity when a normal subject is in a state of mental activity. We measure the synchronous EEG on the auditory event when a subject who is in a normal state thinks of a specific task, and then shift the baseline and reduce the effect of biological artifacts on the measured EEG. Finally we extract only the mental task signal by averaging method, and then perform the recognition of the extracted mental task signal by computing the AR coefficients. In the experiment, the auditory stimulus is used as an event and the EEG was recorded from the three channel $C_3-A_1$, $C_4-A_2$ and $P_Z-A_1$. After averaging 16 times for each channel output, we extracted the features of specific mental tasks by modeling the output as 12th order AR coefficients. We used total 36th order coefficient as an input parameter of the neural network and measured the training data 50 times per each task. With data not used for training, the rate of task recognition is 34-92 percent on the two tasks, and 38-54 percent on the four tasks.