@misc{oai:repo.qst.go.jp:00075759,
 author = {塩川, 桂一郎 and 榊, 泰直 and 西内, 満美子 and 近藤, 康太郎 and ドーバー, ニコラス ピーター and ロウ, ヘーゼル フランシス and 今, 亮 and 渡辺, 幸信 and 神門, 正城 and Shiokawa, Keiichiro and Sakaki, Hironao and Nishiuchi, Mamiko and Kondo, Kotaro and Dover, NicholasPeter and Lowe, HazelFrances and Kon, Akira and Watanabe, Yukinobu and Kando, Masaki},
 month = {Apr},
 note = {Petawatt class laser-solid interaction experiments [1] are conducted at the ultra-short pulse (40fs), ultra-intense (10^22 W/cm^2)  J-KAREN-P laser system[2]. An electron spectrometer (ESM) that can detect up to a maximum electron energy of 30 MeV was used for measurement of the temperature of the hot electron population emitted by the laser-driven plasma. The ESM consists of a 1.0 T magnet, scintillator (DRZ-high) and CCD camera which were placed in the vacuum chamber in order to make real-time measurements. Not only electrons and x-rays emitted by the plasma, but also secondary electron and x-ray emission generated in the vacuum chamber are detected by the ESM. This secondary emission creates background noise which is randomly scattered over the whole of the observed ESM image. 

Recently, Statistical-Learning methods for analysis of “big data” have progressed rapidly allowing a novel denoising technique, known as the Denoising Auto Encoder (DAE) [3], to be established. The DAE is based on a Convolution Neural Network that transforms features extracted from the raw image to produce a processed image in which the unwanted noise component has been digitally removed. This technique has been shown to be suitable for processing the ESM images for the purpose of denoising. In this report, we compare conventional denoising methods with the DAE., HEDS2019},
 title = {Denoising for a real-time electron spectrometer using a Convolution Neural Network},
 year = {2019}
}