论文部分内容阅读
悬浮泥沙浓度是水体监测中极为重要的指标.本论文基于神经网络具有弥补传统经验算法固有误差的潜力,设计并开发了基于人工神经网络的神经网络校正器来对经验反演结果进行二次校正.为了防止在小数据集的情况下出现过拟合问题,采用了特殊设计的正则化项.基于高分五号高光谱遥感数据以及在长江口和沿海水域同时收集的悬浮泥沙浓度实地测量结果,研究了4种基线经验模型,并评估了使用神经网络校正器后的精度.在每个基线模型上都测试了神经网络校正器模型的两个典型应用,包括基线模型校正和时间校正.在这两种应用中,结果均表明,经校正的D′Sa模型具有最高的准确性.通过使用基线模型校正,均方根误差从0.1495 g/L降低至0.1436 g/L,平均绝对百分比误差从0.7821降低至0.7580,决定系数从0.6805升高至0.6926.实施时间校正后,平均绝对百分比误差从0.8657降低至0.7817,决定系数从0.6688升高至0.7155.最后,基于神经网络校正器校正后精度最高的模型处理了整幅高分五号高光谱图像.本论文结果为各种经验反演算法提供了一种通用的二次校正方法,以最大程度地减少基线模型的固有误差,并且保证了反演精度.“,”The suspended sediment concentration(SSC)is an extremely important property for water monitor?ing. Since machine learning technology has been successfully applied in many domains,we combined the strengths of empirical algorithms and the artificial neural network(ANN)to further improve remote sensing re?trieval results. In this study,the neural network calibrator(NNC)based on ANN was proposed to secondarily correct the empirical coarse results from empirical algorithms and generate fine results. A specialized regulariza?tion term has been employed in order to prevent overfitting problem in case of the small dataset. Based on the Gaofen-5(GF-5)hyperspectral remote sensing data and the concurrently collected SSC field measurements in the Yangtze estuarine and coastal waters,we systematically investigated 4 empirical baseline models and evaluated the improvement of accuracy after the calibration of NNC. Two typical applications of NNC models consisting baseline model calibration and temporal calibration have been tested on each baseline models. In both applica?tions,results showed that the calibrated D\'Sa model is of highest accuracy. By employing the baseline model cali?bration,the root mean square error(RMSE)decreased from 0. 1495 g/L to 0. 1436 g/L,the mean absolute per?centage error(MAPE)decreased from 0. 7821 to 0. 7580 and the coefficient of determination(R2)increased from 0. 6805 to 0. 6926. After implementation of the temporal calibration,MAPE decreased from 0. 8657 to 0. 7817 and R2 increased from 0. 6688 to 0. 7155. Finally,the entire GF-5 hyperspectral images on target date were pro?cessed using the NNC calibrated model with the highest accuracy. Our work provides a universal double calibra?tion method to minimize the inherent errors of the baseline models and a moderate improvement of accuracy can be achieved.