Scholarlink

<p>The automation of agriculture is increasingly incorporating advanced technologies like Internet of Things (IoT) and artificial intelligence to optimize productivity. This study focuses on a specific innovation: forecasting post-harvest tomato shelf life using electronic noses and deep learning. The proposed approach provides accurate and continuous monitoring of tomato conditions, addressing the industry’s demand for enhanced efficiency and quality. The model architecture includes four layers: two stacked recurrent layers (either long short-term memory (LSTM) or gated recurrent unit (GRU)), a dropout regularization layer, and a dense output layer. The LSTM based model achieved a mean squared error (MSE) of 0.000586, a mean absolute error (MAE) of 0.01502, and R-squared values of 0.817 (global) and 0.9870 (remaining days). Similarly, the GRU based model demonstrated superior performance with an MSE of 0.000505, an MAE of 0.01392, and R-squared values of 0.832 (global) and 0.9928 (remaining days). By effectively capturing temporal dependencies and minimizing overfitting, these models offer a robust solution for predicting tomato shelf life. This contribution advances agricultural automation, supporting sustainable practices and ensuring the delivery of fresh, high-quality produce to consumers.</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

Forecasting Post-Harvest Tomato Shelf Life with Recurrent Neural Networks and Electronic Nose Data

  • 期刊论文

  • 发表日期：

    2025年8月28日

查看全文
  • Lionel Affognon,
  • Eugène C. Ezin,
  • Chérif Diallo

摘要

The automation of agriculture is increasingly incorporating advanced technologies like Internet of Things (IoT) and artificial intelligence to optimize productivity. This study focuses on a specific innovation: forecasting post-harvest tomato shelf life using electronic noses and deep learning. The proposed approach provides accurate and continuous monitoring of tomato conditions, addressing the industry’s demand for enhanced efficiency and quality. The model architecture includes four layers: two stacked recurrent layers (either long short-term memory (LSTM) or gated recurrent unit (GRU)), a dropout regularization layer, and a dense output layer. The LSTM based model achieved a mean squared error (MSE) of 0.000586, a mean absolute error (MAE) of 0.01502, and R-squared values of 0.817 (global) and 0.9870 (remaining days). Similarly, the GRU based model demonstrated superior performance with an MSE of 0.000505, an MAE of 0.01392, and R-squared values of 0.832 (global) and 0.9928 (remaining days). By effectively capturing temporal dependencies and minimizing overfitting, these models offer a robust solution for predicting tomato shelf life. This contribution advances agricultural automation, supporting sustainable practices and ensuring the delivery of fresh, high-quality produce to consumers.