Hyperspectral remote sensing mineral end member extraction method of twinborn-sparse self-encoding network

• Main Authors: YANG KAIYUAN, WANG MINGWEI, CHENG CHONG, YIN BIYU, WANG CHENG, WU KAIXIONG, YE ZHIWEI, RUAN JINGHOU
• Format: Patent
• Language: Chinese; English
• Published: 29.03.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to application of ensemble learning and a deep neural network in the field of mineral spectrum unmixing, in particular to a hyperspectral remote sensing mineral end member extraction method of a twinborn-sparse self-encoding network. Based on a twin network structure, two groups of sub-networks are used to train multi-modal spectral features at the same time, the similarity of the two sub-networks is compared through weight sharing and updating of the networks, mineral end members in airborne images are corrected in combination with high spectral resolution of ground spectrums, and mineral end member extraction is completed for a research area. According to the method, the feasibility of local space mineral spectrum unmixing is provided, and the defect that a traditional end member extraction method is difficult to visually reflect mineral absorption characteristics is overcome. 本发明涉及集成学习和深度神经网络在矿物光谱解混领域中的应用，尤其是一种孪生-稀疏自编码网络的高光谱遥感矿物端元提取方法。基于孪生网络结构，使用两组子网络对多模态光谱特征同时进行训练，通过网络的权值共享与更新比较二者的相似