Combined PolInSAR And LiDAR Data Study On Inversion Method Of Vegetation Height

The vast forest resources on the earth provide an important ecological guarantee for the rich and diverse species ecology.As the largest miner and master of the earth,human beings should use advanced technology to protect their homes while developing science and technology.Understanding and mastering the forest ecosystem is related to the development plan of a regional ecological environment.In this paper,the inversion method of vegetation height is studied by combining PolInSAR technology with LiDAR sensor data.The content is as follows:(1)The inversion process of UAVSAR single-view composite data collected over Lope National Park and Pongara National Park in Gabon is carried out,and the vegetation heights in the two places are respectively inversed by using the improved RVoG model,and the inversion results are compared with LiDAR data to evaluate.(2)Using SVM algorithm,the vegetation height inversion process of PolInSAR based on the improved RVoG model is added to the multi-baseline selection step,and the baseline selection is transformed into a classification problem.Based on the LiDAR data,the inversion related parameters are gradually weighted and excluded,and then the inversion baseline selection is carried out,so that the PolInSAR inversion height product is related to the LiDAR height product,and a comparative experiment is carried out.Compared with Kapok inversion method,the baseline selection inversion method based on SVM has improved the inversion distribution density standard deviation(SD)by 2.04 and 0.46 in Lope region and Pongara region,respectively,and both of them have achieved good fitting concentration effect.(3)Combining the characteristics of wide range,low cost and poor accuracy of SAR data with the characteristics of narrow range,high cost and high accuracy of LiDAR data,a new multi-sensor data joint sharpening vegetation height inversion process Poliformer is constructed.Using two-way attention mechanism feature extraction network,using self-attention module(SAM)and cross-attention module(CAM)to extract and exchange the height features of the data of UAVSAR sensor and LVIS sensor,and combining the two-way data with simple convolution operation to output a j oint inversion height product.Compared with Kapok algorithm and SVM,the inversion accuracy of Poliformer is compared in two areas,and the performance of fitting curve,standard deviation SD and KS standard value index are evaluated.It is concluded that Poliformer is 0.363 and 0.125 lower than Kapok and SVM in SD index respectively,and the effect of fitting curve is significantly improved,and the model inversion ability is excellent.At the same time,the CAM module structure in the model feature exchange stage is trained and compared,and the standard model structure of three-layer CAM is determined.Finally,the cross-regional inversion ability of Poliformer is studied,and it is found that the inversion adaptability of the model in the corresponding region of untrained data set is weak,and the generalization ability of the model still has great room for improvement.