Scholarlink

<p>Source samples misclassification is a key source for the uncertainty in the aeolian sediment source tracking models. Therefore, identifying sources of aeolian sand and their samples correct classification are necessary to decrease the uncertainty associated with source tracking’s models. This research aimed to introduce deep clustering (DC) as a type of deep learning (DL) models for classifying the source samples for aeolian sand in a region with severe wind erosion and active sand dunes impacting on the railroad networks in the Zhongzaohuo in the Qaidam Basin of Tibetan Plateau, Northwest China. In this research, 40 samples taken from four potential source regions, and then, 40 geochemical elements and elementary compositions were measured in each sample by X-ray fluorescence spectrometer. To classify source samples of aeolian landforms, eight DC models based on deep Autoencoder (DAE) (e.g., Autoencoder, DCN, DEC, IDEC, DipDECK, DipEncoder, DDC and N2D) was employed to classify source samples of aeolian sand in our study area. Then, stepwise discriminant function analysis (DFA) was applied to explore the accuracy of source samples classified correctly in different source classification schemes provided by DC models. The results of the DC models revealed that the accuracy of source samples were classified correctly ranges between 82.5 and 100%. Based on five DC models (e.g., Autoencoder with three-sources (k = 3), DCN (k = 2 and k = 3), Dipencoder (k = 2) and N2D (k = 2), 100% of source samples were classified correctly, whereas according to DipDECK with two-sources (K = 2) as a weakest model, 82.5% of source samples were classified correctly. Overall, DC models based on DAE has high capabilities for the classification and cluster purposes in different fields related to aeolian and fluvial geomorphologies especially for the classification of the source samples for different sediments and deposits such as atmospheric dust, aeolian landforms, loess deposits, and riverine suspended sediment load in the catchments.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Incorporating Deep Clustering Based on Deep Autoencoder into Aeolian Sediment Source Tracking Models

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Baicheng Niu,
  • Mohammad Azim Saraji,
  • Tao Wang,
  • Yougui Song,
  • Lihai Tan,
  • Hamid Gholami

摘要

Source samples misclassification is a key source for the uncertainty in the aeolian sediment source tracking models. Therefore, identifying sources of aeolian sand and their samples correct classification are necessary to decrease the uncertainty associated with source tracking’s models. This research aimed to introduce deep clustering (DC) as a type of deep learning (DL) models for classifying the source samples for aeolian sand in a region with severe wind erosion and active sand dunes impacting on the railroad networks in the Zhongzaohuo in the Qaidam Basin of Tibetan Plateau, Northwest China. In this research, 40 samples taken from four potential source regions, and then, 40 geochemical elements and elementary compositions were measured in each sample by X-ray fluorescence spectrometer. To classify source samples of aeolian landforms, eight DC models based on deep Autoencoder (DAE) (e.g., Autoencoder, DCN, DEC, IDEC, DipDECK, DipEncoder, DDC and N2D) was employed to classify source samples of aeolian sand in our study area. Then, stepwise discriminant function analysis (DFA) was applied to explore the accuracy of source samples classified correctly in different source classification schemes provided by DC models. The results of the DC models revealed that the accuracy of source samples were classified correctly ranges between 82.5 and 100%. Based on five DC models (e.g., Autoencoder with three-sources (k = 3), DCN (k = 2 and k = 3), Dipencoder (k = 2) and N2D (k = 2), 100% of source samples were classified correctly, whereas according to DipDECK with two-sources (K = 2) as a weakest model, 82.5% of source samples were classified correctly. Overall, DC models based on DAE has high capabilities for the classification and cluster purposes in different fields related to aeolian and fluvial geomorphologies especially for the classification of the source samples for different sediments and deposits such as atmospheric dust, aeolian landforms, loess deposits, and riverine suspended sediment load in the catchments.

Graphical Abstract