| Bathymetry is a necessary base for ship navigation, port construction, marine engineering and the island and coastal zone planning. Compared with field bathymetry methods, remote sensing has the advantage of wide coverage, short revisiting period, low cost and high spatial resolution. Especially for the ability of achieving the depth information of the area that is difficult to reach or get close to, therefore, remote sensing is an important means for bathymetry. Because of the shallow reefs that the islands in South China Sea located on, it is difficult for vessels to carry out the measurement, let alone the islands and reefs occupied by other countries. Affected by the wave and sediment conditions, underwater topography of coral reefs changes rapidly. Hence the bathymetry inversion around the islands and reefs in South China Sea with remote sensing is an inevitable choice.To fully utilize the existing resources of remote sensing images and excavate multi-source, multi-angle and multi-temporal information effectively, the paper investigated the fusion models of multi-dimensional remote sensing bathymetry inversion. The study put forward multi-source fusion method of bathymetry inversion based on voting law, and multi-angle and multi-temporal fusion methods of bathymetry inversion based on fuzzy membership degree. Beidao Island of Xisha Islands was chose as the study area, and the accuracy of the methods was evaluated. For the necessary pre-processing step, that is, image noise filtering, this paper analyzed the effect of wavelet noising filtering on the accuracy of depth inversion, explored the impact of different scales of wavelet noise filtering on the accuracy of bathymetry inversion with multi-spectral remote sensing images, and compared the accuracy of wavelet noise filtering with that of traditional denosing methods. The results can be summarized as follows:1. The accuracy of multi-source fusion result of bathymetry inversion was of a considerable improvement as compared to that of single-source inversion result. The average relative error(ARE) dropped significantly with 12.7 percentage points from that of the SPOT-6, whose accuracy was the highest among the single-source inversion results. And the average relative error of the former was 13.1%. Both fusion result and SPOT-6 had the mean absolute error(MAE) of 0.8 m. In the water depth range of 0~2 m, 2~5 m and 5~10 m, as the ARE of multi-source fusion result was less than or equal to the minimum of single-source inversion results, accuracy of fusion result was the highest. The MAE was slightly lower than that of SPOT-6 in 10~20 m.2. The accuracy of multi-angle fusion of bathymetry inversion improved significantly from that of the single-angle inversion, with the ARE of 17.2% and the MAE of 0.8 m, which was 15.3% and 0.1 m lower than those of the forward-angle image separately. The forward-angle image has a higher accuracy in stereo pair. Compared the accuracy of multi-angle fusion result to that of the forward-angle image, the improvement of the former reflected in the depth range of 0~2 m.3. The accuracy of multi-temporal fusion of bathymetry inversion improved slightly from that of the single-temporal inversion, with the ARE of 12.8 lower than that of 2012 Quick Bird inversion result and 1.1% lower than that of 2008 Quick Bird. The MAR decreased by 0.2 m, that is, 1.4 m. To take a view on the accuracy at different water depths, multi-temporal fusion result improved in the depth range of 5~10 m and 10~20 m.4. Wavelet transformation method can maximize the retention of useful information while eliminating noise caused by solar flares, wave breaking and so on to some extent. Compared with the traditional methods of median filtering and mean filtering, it increases the flexibility of filtering window. Therefore, it is a suitable pre-process noise filtering method for bathymetry inversion. |
PDF Full Text Request