Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks
This paper presents a viewpoint from computer vision to the radar echo extrapolation task in the precipitation nowcasting domain. Inspired by the success of some convolutional recurrent neural network models in this domain, including convolutional LSTM, convolutional GRU and trajectory GRU, we desig...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2019-05-01
|
| Series: | Atmosphere |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2073-4433/10/5/244 |
| _version_ | 1819071964311977984 |
|---|---|
| author | Quang-Khai Tran Sa-kwang Song |
| author_facet | Quang-Khai Tran Sa-kwang Song |
| author_sort | Quang-Khai Tran |
| collection | DOAJ |
| description | This paper presents a viewpoint from computer vision to the radar echo extrapolation task in the precipitation nowcasting domain. Inspired by the success of some convolutional recurrent neural network models in this domain, including convolutional LSTM, convolutional GRU and trajectory GRU, we designed a new sequence-to-sequence neural network structure to leverage these models in a realistic data context. In this design, we decreased the numbers of channels in high abstract recurrent layers rather than increasing them. We formulated the task as a problem of encoding five radar images and predicting 10 steps ahead at the pixel level, and found that using only the common mean squared error can misguide the training and mislead the testing. Especially, the image quality of last predictions usually degraded rapidly. As a solution, we employed some visual image quality assessment techniques including Structural Similarity (SSIM) and multi-scale SSIM to train our models. Experimental results show that our structure was more tolerant to increasing uncertainty in the data, and the use of image quality metrics can significantly reduce the blurry image issue. Moreover, we found that using SSIM was very effective and a combination of SSIM with mean squared error and mean absolute error yielded the best prediction quality. |
| first_indexed | 2024-12-21T17:30:12Z |
| format | Article |
| id | doaj.art-53df0c01aace4bb5b426ab6b48b594f8 |
| institution | Directory Open Access Journal |
| issn | 2073-4433 |
| language | English |
| last_indexed | 2024-12-21T17:30:12Z |
| publishDate | 2019-05-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Atmosphere |
| spelling | doaj.art-53df0c01aace4bb5b426ab6b48b594f82022-12-21T18:55:57ZengMDPI AGAtmosphere2073-44332019-05-0110524410.3390/atmos10050244atmos10050244Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural NetworksQuang-Khai Tran0Sa-kwang Song1Department of Big Data Science, University of Science and Technology (UST), Daejeon 34113, KoreaDepartment of Big Data Science, University of Science and Technology (UST), Daejeon 34113, KoreaThis paper presents a viewpoint from computer vision to the radar echo extrapolation task in the precipitation nowcasting domain. Inspired by the success of some convolutional recurrent neural network models in this domain, including convolutional LSTM, convolutional GRU and trajectory GRU, we designed a new sequence-to-sequence neural network structure to leverage these models in a realistic data context. In this design, we decreased the numbers of channels in high abstract recurrent layers rather than increasing them. We formulated the task as a problem of encoding five radar images and predicting 10 steps ahead at the pixel level, and found that using only the common mean squared error can misguide the training and mislead the testing. Especially, the image quality of last predictions usually degraded rapidly. As a solution, we employed some visual image quality assessment techniques including Structural Similarity (SSIM) and multi-scale SSIM to train our models. Experimental results show that our structure was more tolerant to increasing uncertainty in the data, and the use of image quality metrics can significantly reduce the blurry image issue. Moreover, we found that using SSIM was very effective and a combination of SSIM with mean squared error and mean absolute error yielded the best prediction quality.https://www.mdpi.com/2073-4433/10/5/244convolutional LSTMconvolutional GRUtrajectory GRUprecipitation nowcastingradar echo extrapolationimage quality assessment |
| spellingShingle | Quang-Khai Tran Sa-kwang Song Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks Atmosphere convolutional LSTM convolutional GRU trajectory GRU precipitation nowcasting radar echo extrapolation image quality assessment |
| title | Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks |
| title_full | Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks |
| title_fullStr | Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks |
| title_full_unstemmed | Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks |
| title_short | Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks |
| title_sort | computer vision in precipitation nowcasting applying image quality assessment metrics for training deep neural networks |
| topic | convolutional LSTM convolutional GRU trajectory GRU precipitation nowcasting radar echo extrapolation image quality assessment |
| url | https://www.mdpi.com/2073-4433/10/5/244 |
| work_keys_str_mv | AT quangkhaitran computervisioninprecipitationnowcastingapplyingimagequalityassessmentmetricsfortrainingdeepneuralnetworks AT sakwangsong computervisioninprecipitationnowcastingapplyingimagequalityassessmentmetricsfortrainingdeepneuralnetworks |