Multi-modal and multi-objective hyperspectral unmixing model based on multi-source data

Mixed pixels in hyperspectral images are a common imagination in the process of hyperspectral image processing, usually processed using mixed pixel decomposition. To explore the effect of multi-source data fusion on endmember bundle extraction, a multi-modal and multi-objective optimization endmember bundle extraction method based on data fusion (MMO-CDPSO-RSADRDSM) is proposed. Firstly, the initialization particles are randomly assigned in the region by DSM interference; Secondly, according to the characteristics of hyperspectral image space and DSM space, the crowding distance of decision space is improved to improve the diversity of decision space; Finally, the DSM derived weights are integrated into the optimization model to calculate the objective function, and the endmember bundle is extracted iteratively by the above method. The effective endmember bundles extracted by the MMO-CDPSO-RSADRDSM algorithm have mean RMSE (mRMSE) of 0.1853 and 0.1548 on MUUFL and Houston data, respectively, and recombinant minimum mSAD (re-min mSAD) of 0.0325 and 0.0341. The number of endmemebers extracted on the MUUFL dataset is 35, and in Houston it is 36. The experimental results show that this method has good recognition ability for endmembers with high differences and similar hyperspectral bands, and extracts more effective endmember bundles on this basis. This provides a new method for extracting endmember bundles for multi-source data fusion.