Scholarlink

We present in this chapter a strategy for classifying meteorological images using the convolution neural network (CNN). The CNN is applied to images from the Meteosat Second Generation (MSG) meteorological satellites. The study region is divided into several zones considered as considered small image. All zones are assigned to a class, either no rain class or rain class. The processing is carried out on all the images collected during the rainy period from October 2009 to March 2010. The processing consists of introducing these zones extracted from MSG images of different sizes to the CNN and obtaining as output the rain class or the no rain class. The classification results are compared to the results obtained by a standard artificial neural network (ANN). To show the contribution of the CNN model, a comparison of the results with the standard ANN model is performed. The results show better performance in the case of CNN. Indeed, for the POD, for the CNN indicates 95% against 91% for the MLP. The POFD and FAR show 06 and 20% for CNN compared to 09 and 27% for MLP, respectively. The same trends are observed in the case of CSI and PC. 88 and 96% for CNN versus 82 and 92% for MLP, respectively. As for Bias, CNN shows a very slight overestimation which is better than MLP.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of CNN for Separation Between No Rain and Rain from Remote Sensing Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Mourad Lazri,
  • Youcef Attaf,
  • Karim Labadi,
  • Noureddine Bensafi,
  • Slimane Hameg,
  • Soltane Ameur

摘要

We present in this chapter a strategy for classifying meteorological images using the convolution neural network (CNN). The CNN is applied to images from the Meteosat Second Generation (MSG) meteorological satellites. The study region is divided into several zones considered as considered small image. All zones are assigned to a class, either no rain class or rain class. The processing is carried out on all the images collected during the rainy period from October 2009 to March 2010. The processing consists of introducing these zones extracted from MSG images of different sizes to the CNN and obtaining as output the rain class or the no rain class. The classification results are compared to the results obtained by a standard artificial neural network (ANN). To show the contribution of the CNN model, a comparison of the results with the standard ANN model is performed. The results show better performance in the case of CNN. Indeed, for the POD, for the CNN indicates 95% against 91% for the MLP. The POFD and FAR show 06 and 20% for CNN compared to 09 and 27% for MLP, respectively. The same trends are observed in the case of CSI and PC. 88 and 96% for CNN versus 82 and 92% for MLP, respectively. As for Bias, CNN shows a very slight overestimation which is better than MLP.