Key Laboratory of Virtual Geographic Environment, Ministry of Education, Nanjing Normal University, Nanjing 210023, China
The shadow on black and odor water interfere with the spectral information of the water surface and seriously affects the accuracy of water quality monitoring with high spatial resolution remote sensing data. Therefore, it is necessary to remove the shadow before evaluating river water quality. This paper tries to constructan objective and efficient shadow recognition algorithm on black and odor water to reduce the interference of adjacent object and improve the accuracy of remote sensing monitoring and evaluation of river water quality. In this study, the shadow and water pixels were sampled based on the hyperspectral remote sensing data of Unmanned Aerial Vehicle (UAV).The spatial distribution of different band combinations was analyzed by means of spectral feature spatial analysis to obtain spectral band combinations that can effectively distinguish water and water surface shadows, and the coefficients of band combinations were calibrated to obtain the best discrimination effect. By comparing the discernibility of shadow and non-shadow water by various band combinations, it was found that the ration of remote sensing reflectance Rrs(666)/Rrs(791) combining with Rrs(492) has a higher discrimination between water pixels and shadow pixels. Therefore, remote sensing reflectance at 492 nm, 666 nm and 792 nm were selected to establish the River Surface Shadow Index (RSSI). In general, the threshold of distinguishing shadow and non-shadow pixels needs to be adjusted according to different images. In this case, manually adjusting the threshold may produce errors, which are difficult to apply to other images. In order to reduce the error caused by artificial threshold calibration, the maximum category variance method (OTSU)was adopted to automatically determine the threshold of shadow recognition. According to the complexity of the riverbank object, the reflectance spectra of the shadows were classified to two types: umbra and penumbra. The magnitude difference between penumbra and umbra reflectance was similar to that between penumbra and water reflectance. Therefore, in order to highlight the difference between penumbra and water, the number of classification recognition types was set as 3. Firstly, the OTSU method was used to automatically determine the recognition threshold of umbra, penumbra and water, and then the extracted umbra and penumbra were combined to produce the final shadow distribution map. The algorithm was tested by using the hyperspectral remote sensing images of Jinchuan River and Longjiang River in Nanjing. The results show that the RSSI shadow index can highlight the difference between shadow and water. The threshold determined by OTSU adaptively can better distinguish umbra, penumbra and water, and the overall recognition accuracy of shadow can reach more than 85%. This algorithm can effectively identify the water surface shadow on black and odor water and provide the technical support of data preprocessing for the subsequent qualitative and quantitative remote sensing monitoring for water.