A study of automatic measurement of emulsion chamber data

研究了利用图像扫描仪测量高山乳化室实验中观测到的高能宇宙射线族的可能性。结果总结如下：1)用图像扫描仪测量得到的像素值变换后的x射线胶片上的斑点暗度与普通显微光度计测量的暗度有很好的相关性。2)应用图像处理方法，识别雨斑候选点。在这些候选点中，真正的阵雨点是根据几个标准来选择的，这些标准描述了真正的阵雨点，即一个点内局部黑暗的分布。在此过程中，噪声点的数量急剧减少，约为1/(10 ~ 30)。但在四分之一的面积内仍存在约40个噪点；相比之下，真正的阵雨点的数量约为20个。我们还利用人工神经网络在候选点中识别出真实的点，但一些真实的雨点被识别为噪声点，反之亦然。有必要考虑另一个输入参数来表征真正的阵雨点。然而，通过应用以下跟踪方法，可以在大气家族中提取真正的淋浴点和成员淋浴点。3)采用两种不同的方法对连续层的雨斑进行跟踪。一种是利用x射线胶片中雨斑的几何形态信息。如果我们可以手动指定至少一个淋浴轨迹，那么通过比较连续层中每个膜中任意两个淋浴点的相对距离来自动识别其他淋浴轨迹。我们还将hopfield型神经网络应用于人工事件，以研究将其用于寻迹问题的可能性。结果表明，即使存在噪声点和胶片设置不准确的情况下，该网络也能给出基本正确的跟踪，而当存在噪声点时，该网络会给出额外的伪跟踪。少

We have studied a possibility to use image scanner for measuring high energy cosmic-ray families observed by the high mountain emulsion chamber experiments. The results are summarized as follow.1)There are very good correlation between spot darkness in the X-ray films transformed from pixel value obtained by the measurement using image scanner and that measured by using usual micro-photometer.2)Applying image processing method, candidates of shower-spots are identified. Among those candidates spots, true shower-spots are selected according to several criteria which characterize true shower-spots, i.e., a distribution of local darkness inside a spot. The number of noise spots decreases drastically, 〜1/(10〜30) in this procedure. But there still remain 〜40 noise spots in the area of one quarter of the film ; in contrast the number of true shower-spots is 〜20. We also applied artificial neural network to identify the true spots among candidates spots, but some of the true shower-spots were … More identified to be noise spots and vice versa. It is necessary to consider another input parameters to characterize true shower-spots. It is, however, possible to pick up true shower-spots and member shower-spots in a atmospheric family by applying the following tracking method.3)We applied two different methods for tracking of shower-spots in successive layers. One is to use information on geometrical configuration of shower-spots in the X-ray films. If we can manually specify at least one shower track, then the other shower tracks are automatically identified by comparing relative distances of any two shower spots in each film in the successive layers. We applied also the Hopfield-type neural networks to the artificial events in order to study a possibility to use it for the track finding problem. It is found that the networks give practically correct tracking even when there exist noise spots and inaccuracy of film setting, though they give additional fake tracking when noise spots are included. Less