Scholarlink

<p>Machine learning techniques, particularly deep learning algorithms, have emerged as effective tools for identifying mineralization-related geochemical anomalies. Machine learning techniques can be broadly categorized into supervised and unsupervised learning, each with its limitations. Supervised learning relies heavily on large amounts of labeled data, typically using known mineral deposits as positive samples, while the selection of appropriate negative samples poses a challenge. Moreover, it typically fails to explore the valuable information contained in unlabeled data. However, unsupervised learning does not require labeled data but fails to fully utilize the limited available positive samples, leading to lower anomaly detection accuracy. To address these limitations, this paper introduces a deep semi-supervised anomaly detection (DSAD) method for geochemical anomaly identification. This approach trains a model using only known positive samples and unlabeled data, avoiding negative sample selection while effectively leveraging the limited available positive samples. To evaluate the effectiveness of the DSAD method in identifying geochemical anomalies associated with mineralization, the Nanling region in China was selected as a case study. Based on the DSAD framework, two models named AE-DSAD and CAE-DSAD were developed and compared with a deep autoencoder (AE), convolutional autoencoder (CAE), and convolutional neural network (CNN) models. The results indicated that CAE-DSAD performed best in terms of the receiver operating characteristic curve, success-rate curve, and prediction-rate curve, outperforming AE-DSAD, AE, CAE, and CNN. This demonstrates that the CAE-DSAD model exhibits superior anomaly detection capability and provides an effective solution for geochemical anomaly identification.</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

Identification of Geochemical Anomalies Using a Deep Semi-supervised Anomaly Detection Model

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Rui Bi,
  • Di Liu,
  • Qinglin Xia

摘要

Machine learning techniques, particularly deep learning algorithms, have emerged as effective tools for identifying mineralization-related geochemical anomalies. Machine learning techniques can be broadly categorized into supervised and unsupervised learning, each with its limitations. Supervised learning relies heavily on large amounts of labeled data, typically using known mineral deposits as positive samples, while the selection of appropriate negative samples poses a challenge. Moreover, it typically fails to explore the valuable information contained in unlabeled data. However, unsupervised learning does not require labeled data but fails to fully utilize the limited available positive samples, leading to lower anomaly detection accuracy. To address these limitations, this paper introduces a deep semi-supervised anomaly detection (DSAD) method for geochemical anomaly identification. This approach trains a model using only known positive samples and unlabeled data, avoiding negative sample selection while effectively leveraging the limited available positive samples. To evaluate the effectiveness of the DSAD method in identifying geochemical anomalies associated with mineralization, the Nanling region in China was selected as a case study. Based on the DSAD framework, two models named AE-DSAD and CAE-DSAD were developed and compared with a deep autoencoder (AE), convolutional autoencoder (CAE), and convolutional neural network (CNN) models. The results indicated that CAE-DSAD performed best in terms of the receiver operating characteristic curve, success-rate curve, and prediction-rate curve, outperforming AE-DSAD, AE, CAE, and CNN. This demonstrates that the CAE-DSAD model exhibits superior anomaly detection capability and provides an effective solution for geochemical anomaly identification.