NON-DESTRUCTIVE DEEP LEARNING APPROACHES FOR FRUIT QUALITY PREDICTION AND CLASSIFICATION: PAPAYA AND AVOCADO
Date
2024
Authors
Contributor
Advisor
Department
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Narrator
Transcriber
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
Volume
Number/Issue
Starting Page
Ending Page
Alternative Title
Abstract
Artificial intelligence (AI) is a broad discipline that creates systems with abilities to perform a variety of tasks that would normally require human intelligence. Machine learning is a sub-field of artificial intelligence focused on the design of algorithms, giving computers the capability to learn from and analyze decisions based on data. Deep learning is another subset of machine learning that uses many layers of neural networks for modeling complex patterns within big data. Deep learning models are particularly powerful in recognizing patterns and making predictions, making them suitable for a wide range of applications. Consequently, deep learning is applied in various fields, including the food industry. It integrates multiple disciplines and is researched and utilized for various purposes. In the food industry, deep learning holds potential in several aspects, such as quality control and assurance, food safety, and process automation. By leveraging complex neural networks, deep learning can analyze large amounts of data to make accurate predictions and classifications, thus enhancing the efficiency and reliability of processes.Studying and assessing fruit ripening is crucial for ensuring optimal quality, marketability, and consumer satisfaction. This is particularly important for climacteric fruits such as papayas and avocados, which ripen after harvesting. Accurate ripeness assessment can help manage post-harvest handling better, reduce waste, and improve supply chain efficiency.
The study aimed to advance the application of deep learning in non-destructive fruit quality assessment. The main goal is to develop an innovative model that can accurately predict physicochemical properties and classify the ripening stages of fruit. Specifically, the focus is on two models: a CNN (Convolutional Neural Network) that predicts the physicochemical properties of papayas through image analysis and weight evaluation, and a multimodal model that classifies avocado ripening stages using image analysis, acoustic impulse response audio signals, and weight. By integrating these methodologies, the paper seeks to demonstrate the feasibility and efficiency of deep learning techniques to improve fruit quality assessment and provide more accurate, non-destructive and automated solutions to the agricultural sector.
In the first study (Chapter 3), images of papaya at different stages of ripeness were captured and weighed. From a total of 132 papayas, 528 images were obtained. Through image augmentation, the number of images increased threefold to 1584 images, which were used in the study. Various physicochemical properties such as texture, pH, total soluble solids (TSS), and seed weight were subsequently measured. The data were split into training, validation, and test sets in an 8:1:1 ratio for model training and evaluation. A CNN model was trained using image snapshots and weights as input values to predict physicochemical property values. Model performance was evaluated using the mean squared error (MSE) and the coefficient of determination (R²) as evaluation metrics. The CNN model achieved the high accuracy with MSE values of 0.0284 and 0.1729 and R² values ranging from 0.71 to 0.94 for the training and validation sets, respectively. These findings demonstrated that CNN-based models can provide detailed and quantitative insights to improve the understanding and management of papaya quality and characteristics.
The second study (Chapter 4) involved a multimodal approach to classify Hass avocado ripening stages (unripe, ripe, or overripe) using image analysis, audio analysis, and weight measurements. Data points, including 584 images, acoustic impulse response audio signals, and weight measurements, were collected from 146 Hass avocados. The image data underwent image augmentation, resulting in the image data being doubled to 1168 images. Additionally, the audio data were converted to log-Mel spectrograms, and the avocado firmness data were processed using Box-Cox transformation. The collected data were split into training, validation, and test sets in an 8:1:1 ratio for model training and evaluation. A multimodal ripeness classification model integrating image, audio, and weight (i.e. IAWM) was used to predict avocado firmness and classify ripeness. Model performance was validated using accuracy, precision, recall, F1 score, and confusion matrix. The IAWM model was compared to the image-based ripeness classification model (IM) and multimodal ripeness classification model integrating image and audio (IAM). The IAWM achieved an accuracy of 0.949, compared to 0.838 for the IM and 0.897 for the IAM. In addition, the IAWM demonstrated superior performances in terms of the precision, recall, and F1 score. These results indicate that integrating multiple data sources improves the ability of the deep learning model to accurately classify ripeness stages.
Both studies show that non-destructive deep learning approaches can significantly enhance fruit quality assessment. The CNN model for papayas and the multimodal model for avocados demonstrate the potential of deep learning to offer accurate, efficient, and practical solutions for predicting physicochemical properties and classifying ripeness stages. These advances promise to improve quality control, reduce waste and increase consumer satisfaction in the agricultural sector. Unified data formats not only improve fruit quality predictive modeling, but also help gain insights and better management formats.
This study demonstrates the potential of deep learning approaches in non-destructive fruit quality assessment. By optimizing the parameters of the model and integrating various data sources, the accuracy and efficiency of fruit quality prediction and classification can be significantly improved, providing substantial benefits to the food industry.
Description
Keywords
Food science, Food science
Citation
Extent
154 pages
Format
Geographic Location
Time Period
Related To
Related To (URI)
Table of Contents
Rights
All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.
Rights Holder
Local Contexts
Collections
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.