Scholarlink

Fruits, especially bananas, are one of the world’s most popular and widely consumed fruits, known for their nutritional value and convenience. However, achieving optimal consumer satisfaction and reducing supply chain wastage requires accurate assessment of banana ripeness. Accurately determining banana maturity is crucial for optimizing harvest and distribution while preserving the right vitamins or substances. Current methods rely heavily on manual judgment, making them subjective, inconsistent, and labor-intensive. Therefore, a more efficient and consistent solution is needed to determine banana ripeness. Hyperparameters that are essential for neural network deployment and tuning these hyperparameters can greatly affect the convergence speed and the quality of the learned model. In this work, the focus was on fine-tuning hyperparameters such as number of batch size, number of epochs, type of optimizer, rectified unit and its evaluation for classification of bananas ripeness. The dataset used is the Banana Ripeness Images dataset, which consists of 4 classes, each containing 2011 images, totaling 8044 images. The dataset is divided into two parts: 80% for the training dataset and 20% for the validation dataset. For testing, randomly selected images are taken from Google for each class. The results show that the highest accuracy value for the training model is 99.63% when using the RELU activation function, NADAM optimizer, a batch size of 64, and 20 epochs. When the model is tested with the testing dataset, it is found that the accuracy is 100% for all categories.

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

A Fine-Tuning Hyperparameters for Performance Comparison of Fruit Ripeness Classification Using a Convolutional Neural Network

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Abd Rasid Mamat,
  • Muhammad Asif Affan Zulkifli,
  • Mohamad Afendee Mohamed,
  • Norkhairani Abdul Rawi,
  • Mohd Fadzil Abd Kadir,
  • Azim Zaliha Abd Aziz,
  • Mohd Khalid Awang

摘要

Fruits, especially bananas, are one of the world’s most popular and widely consumed fruits, known for their nutritional value and convenience. However, achieving optimal consumer satisfaction and reducing supply chain wastage requires accurate assessment of banana ripeness. Accurately determining banana maturity is crucial for optimizing harvest and distribution while preserving the right vitamins or substances. Current methods rely heavily on manual judgment, making them subjective, inconsistent, and labor-intensive. Therefore, a more efficient and consistent solution is needed to determine banana ripeness. Hyperparameters that are essential for neural network deployment and tuning these hyperparameters can greatly affect the convergence speed and the quality of the learned model. In this work, the focus was on fine-tuning hyperparameters such as number of batch size, number of epochs, type of optimizer, rectified unit and its evaluation for classification of bananas ripeness. The dataset used is the Banana Ripeness Images dataset, which consists of 4 classes, each containing 2011 images, totaling 8044 images. The dataset is divided into two parts: 80% for the training dataset and 20% for the validation dataset. For testing, randomly selected images are taken from Google for each class. The results show that the highest accuracy value for the training model is 99.63% when using the RELU activation function, NADAM optimizer, a batch size of 64, and 20 epochs. When the model is tested with the testing dataset, it is found that the accuracy is 100% for all categories.