Scholarlink

<p>To improve the recognition accuracy of acoustic vibration signals, this study proposed an intelligent method for detecting moldy core in Korla pear (Pyrus sinkiangensis) based on time–frequency dual-channel feature fusion and attention enhancement. Response signals from pears with different health conditions were collected using a self-designed acoustic vibration detection system. In the time-domain branch, a temporal convolutional network (TCN) was used to capture temporal dependencies in the acoustic vibration signals. In the frequency-domain branch, Discrete Wavelet Transform (DWT) was combined with a convolutional neural network (CNN) to extract multi-scale spectral features. Subsequently, a one-dimensional enhanced convolutional block attention module (1DECBAM) was introduced to strengthen feature representation, and a learnable weighted fusion mechanism was adopted to adaptively integrate the time- and frequency-domain features, thereby enabling information interaction and complementarity between the two feature domains. To identify pears with different degrees of moldy core, three classifiers, namely XGBoost, MLP, and Softmax, were constructed, and their classification performance based on fused features was compared. The experimental results showed that the proposed TF-AXGNet model achieved an accuracy of 95.95 ± 0.35% in the three-class identification task. These results demonstrate that the proposed method can enable early, rapid, and nondestructive detection of moldy core in Korla pears, providing a feasible technical approach for the intelligent detection of internal fruit diseases.</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

Non-destructive identification of moldy pear core using acoustic vibration analysis and dual-channel deep feature fusion

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Kang Zhao,
  • Xuezhen Wang,
  • Yue Yang,
  • Shuhui Bi,
  • Han Zhang,
  • Tao Shen

摘要

To improve the recognition accuracy of acoustic vibration signals, this study proposed an intelligent method for detecting moldy core in Korla pear (Pyrus sinkiangensis) based on time–frequency dual-channel feature fusion and attention enhancement. Response signals from pears with different health conditions were collected using a self-designed acoustic vibration detection system. In the time-domain branch, a temporal convolutional network (TCN) was used to capture temporal dependencies in the acoustic vibration signals. In the frequency-domain branch, Discrete Wavelet Transform (DWT) was combined with a convolutional neural network (CNN) to extract multi-scale spectral features. Subsequently, a one-dimensional enhanced convolutional block attention module (1DECBAM) was introduced to strengthen feature representation, and a learnable weighted fusion mechanism was adopted to adaptively integrate the time- and frequency-domain features, thereby enabling information interaction and complementarity between the two feature domains. To identify pears with different degrees of moldy core, three classifiers, namely XGBoost, MLP, and Softmax, were constructed, and their classification performance based on fused features was compared. The experimental results showed that the proposed TF-AXGNet model achieved an accuracy of 95.95 ± 0.35% in the three-class identification task. These results demonstrate that the proposed method can enable early, rapid, and nondestructive detection of moldy core in Korla pears, providing a feasible technical approach for the intelligent detection of internal fruit diseases.

Graphical abstract