Sonal B. Nikam, Sudhir B. Lande, Vinay J. Nagalkar, G. C. Wakchaure, Paramasivam Suresh Kumar
{"title":"基于融合深度CNN特征和集成学习的火龙果质量分级和成熟度预测分类模型","authors":"Sonal B. Nikam, Sudhir B. Lande, Vinay J. Nagalkar, G. C. Wakchaure, Paramasivam Suresh Kumar","doi":"10.1155/jfpp/6938071","DOIUrl":null,"url":null,"abstract":"<p><b>Introduction:</b> Dragon fruit quality grading and maturity detection are crucial for ensuring market value and consumer satisfaction. The manual grading and maturity assessment of dragon fruit is a tedious, time-consuming task that often lacks consistency due to human subjectivity and environmental factors. Existing computer vision–based classification models rely on single deep learning architectures, which may not effectively capture the diverse features required for accurate quality assessment. Moreover, traditional machine learning classifiers struggle to generalize well when dealing with complex agricultural datasets containing variations in fruit texture, size, and lighting conditions.</p><p><b>Method:</b> This paper presents a predictive classification model that integrates fused deep convolutional neural network (CNN) features with ensemble learning to achieve high-accuracy classification. The proposed model extracts deep features from multiple pretrained CNN architectures, including VGG16 and Inception-ResNet-v2, and fuses them to enhance feature representation. These fused deep features were then classified using an ensemble of machine learning algorithms, including random forest (RF), gradient boosting (GB), and AdaBoost (AB).</p><p><b>Findings:</b> Experimental results on a benchmark dataset of dragon fruit images demonstrated that the proposed hybrid approach outperformed individual deep learning and machine learning models, achieving superior accuracy, precision, and recall in quality and maturity grade classification. RF achieved 99.99% accuracy in quality grading by classifying fresh versus defected fruits, while AB reached 99.69% accuracy in maturity detection, surpassing the performance of both RF and GB.</p><p><b>Conclusions:</b> The model effectively distinguished different maturity levels and quality grades, offering a robust and automated solution for dragon fruit quality assessment. Its high accuracy and adaptability support real-world deployment in smart farming and processing units, including in automated sorting and grading systems. This enables real-time, consistent, and accurate fruit classification, thereby reducing labor, minimizing human error, and improving supply chain efficiency.</p>","PeriodicalId":15717,"journal":{"name":"Journal of Food Processing and Preservation","volume":"2025 1","pages":""},"PeriodicalIF":2.5000,"publicationDate":"2025-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1155/jfpp/6938071","citationCount":"0","resultStr":"{\"title\":\"Predictive Classification Model for Quality Grading and Maturity Detection of Dragon Fruit Using Fused Deep CNN Feature and Ensemble Learning\",\"authors\":\"Sonal B. Nikam, Sudhir B. Lande, Vinay J. Nagalkar, G. C. Wakchaure, Paramasivam Suresh Kumar\",\"doi\":\"10.1155/jfpp/6938071\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><b>Introduction:</b> Dragon fruit quality grading and maturity detection are crucial for ensuring market value and consumer satisfaction. The manual grading and maturity assessment of dragon fruit is a tedious, time-consuming task that often lacks consistency due to human subjectivity and environmental factors. Existing computer vision–based classification models rely on single deep learning architectures, which may not effectively capture the diverse features required for accurate quality assessment. Moreover, traditional machine learning classifiers struggle to generalize well when dealing with complex agricultural datasets containing variations in fruit texture, size, and lighting conditions.</p><p><b>Method:</b> This paper presents a predictive classification model that integrates fused deep convolutional neural network (CNN) features with ensemble learning to achieve high-accuracy classification. The proposed model extracts deep features from multiple pretrained CNN architectures, including VGG16 and Inception-ResNet-v2, and fuses them to enhance feature representation. These fused deep features were then classified using an ensemble of machine learning algorithms, including random forest (RF), gradient boosting (GB), and AdaBoost (AB).</p><p><b>Findings:</b> Experimental results on a benchmark dataset of dragon fruit images demonstrated that the proposed hybrid approach outperformed individual deep learning and machine learning models, achieving superior accuracy, precision, and recall in quality and maturity grade classification. RF achieved 99.99% accuracy in quality grading by classifying fresh versus defected fruits, while AB reached 99.69% accuracy in maturity detection, surpassing the performance of both RF and GB.</p><p><b>Conclusions:</b> The model effectively distinguished different maturity levels and quality grades, offering a robust and automated solution for dragon fruit quality assessment. Its high accuracy and adaptability support real-world deployment in smart farming and processing units, including in automated sorting and grading systems. This enables real-time, consistent, and accurate fruit classification, thereby reducing labor, minimizing human error, and improving supply chain efficiency.</p>\",\"PeriodicalId\":15717,\"journal\":{\"name\":\"Journal of Food Processing and Preservation\",\"volume\":\"2025 1\",\"pages\":\"\"},\"PeriodicalIF\":2.5000,\"publicationDate\":\"2025-09-25\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://onlinelibrary.wiley.com/doi/epdf/10.1155/jfpp/6938071\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Food Processing and Preservation\",\"FirstCategoryId\":\"97\",\"ListUrlMain\":\"https://onlinelibrary.wiley.com/doi/10.1155/jfpp/6938071\",\"RegionNum\":3,\"RegionCategory\":\"农林科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"FOOD SCIENCE & TECHNOLOGY\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Food Processing and Preservation","FirstCategoryId":"97","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1155/jfpp/6938071","RegionNum":3,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"FOOD SCIENCE & TECHNOLOGY","Score":null,"Total":0}
Predictive Classification Model for Quality Grading and Maturity Detection of Dragon Fruit Using Fused Deep CNN Feature and Ensemble Learning
Introduction: Dragon fruit quality grading and maturity detection are crucial for ensuring market value and consumer satisfaction. The manual grading and maturity assessment of dragon fruit is a tedious, time-consuming task that often lacks consistency due to human subjectivity and environmental factors. Existing computer vision–based classification models rely on single deep learning architectures, which may not effectively capture the diverse features required for accurate quality assessment. Moreover, traditional machine learning classifiers struggle to generalize well when dealing with complex agricultural datasets containing variations in fruit texture, size, and lighting conditions.
Method: This paper presents a predictive classification model that integrates fused deep convolutional neural network (CNN) features with ensemble learning to achieve high-accuracy classification. The proposed model extracts deep features from multiple pretrained CNN architectures, including VGG16 and Inception-ResNet-v2, and fuses them to enhance feature representation. These fused deep features were then classified using an ensemble of machine learning algorithms, including random forest (RF), gradient boosting (GB), and AdaBoost (AB).
Findings: Experimental results on a benchmark dataset of dragon fruit images demonstrated that the proposed hybrid approach outperformed individual deep learning and machine learning models, achieving superior accuracy, precision, and recall in quality and maturity grade classification. RF achieved 99.99% accuracy in quality grading by classifying fresh versus defected fruits, while AB reached 99.69% accuracy in maturity detection, surpassing the performance of both RF and GB.
Conclusions: The model effectively distinguished different maturity levels and quality grades, offering a robust and automated solution for dragon fruit quality assessment. Its high accuracy and adaptability support real-world deployment in smart farming and processing units, including in automated sorting and grading systems. This enables real-time, consistent, and accurate fruit classification, thereby reducing labor, minimizing human error, and improving supply chain efficiency.
期刊介绍:
The journal presents readers with the latest research, knowledge, emerging technologies, and advances in food processing and preservation. Encompassing chemical, physical, quality, and engineering properties of food materials, the Journal of Food Processing and Preservation provides a balance between fundamental chemistry and engineering principles and applicable food processing and preservation technologies.
This is the only journal dedicated to publishing both fundamental and applied research relating to food processing and preservation, benefiting the research, commercial, and industrial communities. It publishes research articles directed at the safe preservation and successful consumer acceptance of unique, innovative, non-traditional international or domestic foods. In addition, the journal features important discussions of current economic and regulatory policies and their effects on the safe and quality processing and preservation of a wide array of foods.