Scholarlink

<p>Fruit hardness is an important feature related to maturity level, its accurate estimation is of great significance to the intelligent fruit grading industry. This paper integrates tactile sensing technology with advanced deep learning algorithms to propose an innovative non-destructive method for assessing kiwifruit hardness. Firstly, a haptic information acquisition system was built to verify the performance of the haptic sensor array, collect multidimensional time series data, and conduct pre-processing and dimensional correlation analysis of the data. Then, a new network framework combining graph learning, graph convolutional neural network (GCN) and hierarchical graph pooling is designed. This framework enables real-time modeling of spatial correlations and temporal dependencies within tactile sensor arrays while preserving critical hardness-sensitive features. This addresses the shortcomings of existing temporal modeling methods based on Graph Neural Network (GNN), which tend to overlook cross-sensor correlations or lose valuable information during the pooling process. The results show that this framework accurately distinguishes the hardness levels of three different ripeness stages of kiwifruit, with the highest assessment accuracy of 99.59%. The proposed method was validated through comparative experiments and ablation studies, while its feature capture capability was verified by visualization method. Finally, the nondestructive properties of the method were verified by hardness test and breath test. This study offers a promising solution for non-destructive hardness evaluation of kiwifruit and holds potential for integration into automated production lines to achieve higher-precision intelligent sorting.</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

Nondestructive evaluation of Kiwi hardness by fusion of tactile sensing array and deep learning

  • 期刊论文

  • 发表日期：

    2026年2月7日

查看全文
  • Lihua Cai,
  • Xingyu Hu,
  • Jinke Gao,
  • Hongyao Chen,
  • Haifeng Fang

摘要

Fruit hardness is an important feature related to maturity level, its accurate estimation is of great significance to the intelligent fruit grading industry. This paper integrates tactile sensing technology with advanced deep learning algorithms to propose an innovative non-destructive method for assessing kiwifruit hardness. Firstly, a haptic information acquisition system was built to verify the performance of the haptic sensor array, collect multidimensional time series data, and conduct pre-processing and dimensional correlation analysis of the data. Then, a new network framework combining graph learning, graph convolutional neural network (GCN) and hierarchical graph pooling is designed. This framework enables real-time modeling of spatial correlations and temporal dependencies within tactile sensor arrays while preserving critical hardness-sensitive features. This addresses the shortcomings of existing temporal modeling methods based on Graph Neural Network (GNN), which tend to overlook cross-sensor correlations or lose valuable information during the pooling process. The results show that this framework accurately distinguishes the hardness levels of three different ripeness stages of kiwifruit, with the highest assessment accuracy of 99.59%. The proposed method was validated through comparative experiments and ablation studies, while its feature capture capability was verified by visualization method. Finally, the nondestructive properties of the method were verified by hardness test and breath test. This study offers a promising solution for non-destructive hardness evaluation of kiwifruit and holds potential for integration into automated production lines to achieve higher-precision intelligent sorting.