International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 09 | Sep 2025
p-ISSN: 2395-0072
www.irjet.net
FreshIQNet: A Deep Learning-Powered Framework for Real-Time Multiclass Fruit Quality Detection and Classification Prof.Savita S G1, Mahesh2 1Professor,Master of Computer Application, VTU CPGS, Kalaburagi, Karnataka, India 2 Student, Master of Computer Application, VTU CPGS, Kalaburagi, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract- Fruit quality plays a crucial role in consumer
extraction capabilities, excelling in image classification, transfer learning, and real-time applications. This study presents FreshIQNet, a deep learning framework for realtime multiclass fruit quality detection and classification. Using InceptionResNetV2 as the backbone and integrating custom dense and dropout layers, the system categorizes fruits—including apples, bananas, guavas, lemons, limes, oranges, and pomegranates—into classes such as Good, Bad, and Mixed. Preprocessing and augmentation techniques, along with stratified train-test splitting, enhance generalization and prevent overfitting. The framework delivers real-time predictions via a web interface, enabling image uploads, instant classification, and automated logging for monitoring and analytics. By minimizing human error, improving throughput, and providing actionable insights, FreshIQNet offers a scalable, explainable solution for automated fruit quality monitoring and establishes a foundation for future precision agriculture and smart supply chain applications.
health, market value, and supply chain efficiency, yet conventional inspection methods are manual, subjective, and inconsistent. To address these limitations, this study introduces FreshIQNet, a deep learning-based framework for real-time multiclass fruit quality detection and classification. The model employs InceptionResNetV2 as a feature extractor with a customized classification head to categorize multiple fruit types—apples, bananas, guavas, lemons, limes, oranges, and pomegranates—into distinct quality classes such as Good, Bad, and Mixed. Robust preprocessing techniques, data augmentation, and stratified train–test splitting enhance model generalization and stability. Experimental results demonstrate high classification accuracy, supported by confusion matrices and class-wise probability analysis for explainability. A user-friendly web interface enables image uploads, live predictions, and automated logging, facilitating seamless deployment. FreshIQNet highlights the potential of deep convolutional neural networks to transform fruit quality assessment into intelligent, scalable, and automated systems for agriculture, retail, and warehouse applications.
2. RELATED WORKS Article [1] 'Automatic fruit classification using InceptionResNetV2' by S. G. et al. in 2019: This paper targets supermarket and retail use-cases where automatic fruit recognition reduces checkout errors and speeds handling, adopting Inception-ResNetV2 for robust visual feature extraction from fruit images. The study positions deep transfer learning as a superior alternative to manual features for multiclass fruit classification across varied lighting and background conditions. It discusses input preprocessing and data normalization steps that stabilize training of deep residual-inception hybrids on consumer-grade datasets. The authors compare baseline CNNs with Inception-ResNetV2, noting improved top-1 accuracy and better generalization on unseen classes. Ablations indicate that mixed residual connections help preserve gradient flow in deeper layers for texture and color cues crucial to fruit identification.
Keywords: Fruit quality assessment, Deep learning, InceptionResNetV2, Real-time classification, agriculture, Computer vision, Food technology.
Smart
1.INTRODUCTION Fruits are a vital part of a balanced diet, providing essential vitamins, minerals, antioxidants, and dietary fiber, and their quality directly influences consumer satisfaction, nutritional value, and economic outcomes in agriculture and the food supply chain. Traditionally, fruit quality assessment has relied on manual inspection, where experts evaluate physical attributes such as color, size, texture, ripeness, and visible defects. However, these methods are labor-intensive, subjective, time-consuming, and prone to human error, limiting their applicability for large-scale or real-time evaluation. The rising demand for fresh, high-quality fruits in global markets has motivated research into automated, intelligent systems for accurate fruit quality assessment. Advances in computer vision and deep learning, particularly Convolutional Neural Networks (CNNs), have enabled automated extraction of hierarchical features from images, reducing the need for manual feature engineering. Pretrained models like InceptionResNetV2 offer robust feature
© 2025, IRJET
|
Impact Factor value: 8.315
Article[2]'Fruit classification using attention-based MobileNetV2 for industrial applications' by T. B. Shahi et al. in 2022: The authors propose a lightweight MobileNetV2 enhanced with attention modules to achieve high accuracy with low FLOPs, suitable for real-time systems on embedded hardware. Extensive experiments show the attentionaugmented network improves discriminative focus on
|
ISO 9001:2008 Certified Journal
|
Page 413