Amberlytics: AI-Powered Mango Classification, Grading &

Pricing

Aditya Rajendra Mishra Prem Prabhakar Patil

Information Technology Information Technology
Saraswati College Of Engineering Saraswati College Of Engineering
[email protected] [email protected]

Sweta Madan Shah Chaitanya Ashok Shivade

Information Technology Information Technology
Saraswati College Of Engineering Saraswati College Of Engineering
[email protected] [email protected]

Prof. Maahi Khemchandani

Information Technology
Saraswati College Of Engineering
[email protected]

Saraswati College of Engineering, Navi Mumbai, Maharashtra, India

Abstract- mango varieties has useful applications in agriculture, trade,

and quality control. Traditional mango classification relies on
Mango classification is crucial in agriculture, distribution, human experts or manual measurements of morphological
and consumer markets, as misidentification can lead to features. However, this can be slow, subjective and error-
quality control issues, pricing inconsistencies, and unmet prone. With advances in computer vision and deep learning,
consumer expectations. To address these challenges, we automated mango classification from images is now feasible.
propose a web application integrated with a pre-trained Convolutional neural networks (CNNs) in particular have
convolutional neural network (CNN) model for real-time achieved impressive results on complex visual recognition
mango species identification. The system also incorporates tasks. The global diversity of mango species presents a
price prediction and grading functionalities, categorizing challenge for importers, who must ensure they receive the
mangoes into three quality grades to enhance quality exact species they’ve ordered when importing man-goes from
assurance and ensure pricing transparency. different regions. To address this issue, there’s a growing need
Beyond domestic markets, this solution benefits foreign fora reliable classification system that can accurately
importers by simplifying the identification of diverse differentiate between various mango species. Such a system
Indian mango varieties, ensuring smoother transactions would provide importers with a valuable tool to verify their
and building confidence in international trade. By products and avoid any potential confusion or
leveraging deep learning techniques, the application aims misidentification when dealing with international mango
to standardize mango classification, support better trade.
decision-making in the supply chain, and improve overall
consumer satisfaction. II. LITERATURE SURVEY

Keywords- Mango classification, CNN model, quality Numerous studies have explored automated mango
grading, price prediction, international trade. classification and grading using deep learning, machine
vision, and machine learning techniques. These approaches
aim to improve efficiency in agro-industries by reducing
manual labor and increasing accuracy in identifying mango
I. INTRODUCTION varieties, ripeness, and quality.
Mango is an economically important fruit crop grown in
One study developed an automated mango classification
tropical and subtropical regions around the world. India is the
system using deep learning, particularly convolutional neural
largest producer of mangoes, accounting for over 40% of
networks (CNNs) with transfer learning. The Inception V3
global production. There are numerous different mango
model, pretrained on ImageNet, was retrained on a dataset of
varieties grown, with varying shapes, sizes, colors , flavours
2000 images covering 12 mango varieties, achieving an
and textures. Automatic classification and identification of
accuracy of 97.6%. Similarly, another study used GoogleNet
CNN architecture on 5093 mango images belonging to five Existing systems for mango classification, grading, and price
Indian mango varieties, achieving a test accuracy of 99%. prediction rely on machine learning (ML), deep learning (DL),
These studies demonstrate the effectiveness of CNNs and image processing techniques. Traditional manual
combined with transfer learning in improving classification inspection methods based on size, color, and texture were
accuracy. However, they highlight the need for larger and subjective and inconsistent. Price estimation also depended on
more diverse datasets to enhance model robustness. expert judgment rather than data-driven models.

A machine vision-based system was proposed for predicting Modern ML-based systems use machine vision to automate
the maturity of harvested mangoes using Gaussian Mixture mango inspection. Feature-based models like Support Vector
Models (GMM). The system extracted features such as size, Machines (SVM), Random Forest (RF), and k-Nearest
shape, color, and texture, achieving over 90% accuracy in Neighbors (KNN) classify mangoes based on extracted
classifying mango maturity levels. While this approach features. Deep learning models, especially Convolutional
enables non-destructive ripeness evaluation, it requires Neural Networks (CNNs) such as Inception V3, ResNet, and
controlled lighting conditions for consistent results. Another MobileNet, achieve high classification accuracy (up to 99%).
study combined machine vision and artificial intelligence for However, deep learning requires large datasets and significant
mango sorting, grading, and quality evaluation. It used image computational power, limiting real-time applications.
processing to detect mango shape, size, and surface defects,
alongside a neural network model that predicted maturity and Price prediction models use regression techniques based on
sweetness with 98% accuracy. However, testing was limited fruit attributes and market trends. While linear regression
to a few Vietnamese mango varieties, and lighting provides reasonable accuracy, it struggles with sudden market
inconsistencies posed challenges. fluctuations. Advanced deep learning models like Long Short-
Term Memory (LSTM) networks improve price forecasting
Embedded machine learning techniques were explored for accuracy, achieving a mean absolute percentage error (MAPE)
mango weight estimation using image processing and Support of just 3.5%.
Vector Machines (SVM) on a Raspberry Pi platform. This
system categorized mangoes into three weight classes with Despite advancements, current systems face challenges like
over 85% accuracy. While cost-effective for developing data limitations, poor performance under variable lighting, and
countries, its real-time application was hindered by slow high computational requirements. Existing price prediction
processing times. Another study focused on mango disease models lack real-time market integration, reducing reliability.
detection, specifically anthracnose fungal infection, using an Future systems like MangoSpecs should combine deep
8-layer CNN model. The model achieved 96.16% accuracy, learning, advanced imaging, and AI-driven market analysis to
demonstrating the potential of deep learning in plant disease enhance classification, grading, and price prediction accuracy
detection. However, its evaluation was limited to a single while improving real-time processing and scalability.
disease type, necessitating further validation across multiple
mango diseases.

Beyond mangoes, researchers have applied deep learning for IV. PROPOSED SYSTEM
plant classification, fish species detection, and plant disease
identification. A study on tree species classification used a The proposed system, MangoSpecs, is an intelligent solution
CNN model on 10,000 leaf images from 10 tree species, for mango species classification, price prediction, and quality
achieving 99.4% accuracy. Another study employed Faster R- grading using a Convolutional Neural Network (CNN). The
CNN for detecting five plant diseases with over 95% model is trained on a diverse dataset containing multiple
accuracy, optimized for mobile deployment. These findings mango species, including Kesar, Dasheri, Chaunsa, Langda,
suggest the broader applicability of deep learning in Alphonso (Hapus), and Totapuri. The dataset is split into
agriculture. training and testing subsets to ensure accurate model
development and validation.
Mango grading systems have also been explored using CNNs,
with one study comparing VGG16, ResNet152, and Inception To make the system accessible and user-friendly, a web-based
V3 architectures for classifying eight mango cultivars. application is integrated with the pretrained CNN model.
Inception V3 outperformed the others, achieving 99.2% Users can upload mango images to the web app, which
classification accuracy and 96.7% grading accuracy. Another processes the image and provides real-time identification
study employed image processing to classify mango ripeness results, including:
based on color variations, using different color spaces and
machine learning classifiers.  Mango species classification
 Minimum and maximum pricing based on historical
Overall, deep learning and machine vision offer promising data and market trends
solutions for mango classification and grading. However,  Quality grading, categorizing the mango into
challenges such as dataset limitations, inconsistent lighting, predefined quality classes
and the need for real-time efficiency must be addressed to
enable industrial-scale adoption. Future research should focus The system leverages deep learning and image processing
on expanding datasets, improving model robustness, and techniques to ensure high accuracy in mango classification
integrating multi-sensor imaging techniques for enhanced and quality assessment. Transfer learning may be applied to
accuracy. enhance model performance while reducing the need for
extensive labeled datasets. Additionally, the system can be
expanded in the future to incorporate more mango varieties,
III. EXISTING SYSTEM
improve real-time price estimation, and refine quality grading The computed results—mango variety, confidence level,
criteria. grade, ripeness category, and price range—are formatted into
JSON and sent back to the frontend. Users receive the output
The MangoSpecs system aims to assist farmers, traders, and in a structured format, enabling them to make informed
consumers in making informed decisions regarding mango decisions regarding their mango's classification and potential
classification, pricing, and quality, thereby enhancing market value.
efficiency in the agro-based industry.

V. METHODOLOGY

1. Image Acquisition

Users interact with a web-based interface, uploading mango

images via drag-and-drop or file selection. Additionally, they
specify their city, which is used to adjust price calculations
based on local market conditions. This step ensures that the
system receives user-provided image data while capturing
location-based pricing factors.

2. Preprocessing

Upon receiving an image, the system performs preprocessing

operations, including resizing it to 128x128 pixels and
normalizing pixel values to a range of [0,1]. These steps
standardize input images for compatibility with the
Convolutional Neural Network (CNN) model. By enforcing a
uniform input format, the model can deliver consistent and
accurate classification results.

3. Mango Classification

A deep learning-based CNN model is employed to classify the

mango into one of 15 predefined varieties, such as Alphonso
or Kesar. The classification process is driven by a model
trained on an extensive dataset, ensuring high accuracy (95%
+). The system selects the mango variety with the highest
probability score and applies any predefined post-
classification corrections.

4. Quality and Ripeness Assessment

Based on the confidence level of the classification result, the

mango is assigned a grade (A, B, or C). Additionally, ripeness
is determined by confidence thresholds, categorizing the
mango as Ripe, Almost Ready, or Unripe. This ensures that
both the quality and maturity of the mango are accounted for
in the final analysis.

5. Price Estimation

The system retrieves the base price range for the classified
mango variety. A city-specific price adjustment factor is
applied, modifying the price estimate according to regional
market trends. The final price range is computed dynamically,
reflecting real-time economic variations and local demand. Fig.1 Workflow of the model

6. Response Generation & Display

V. RESULTS
The results obtained from the trained model showcase its
ability to accurately differentiate between different varieties
mango varieties based on image 39MangoSpecs inputs and
segregate them into different classes i.e. Grade 1 (extremely
good/export quality), Grade 2(with some spots or marks),
Grade 3(not acceptable).
Further the model also describes the price of the said specie of
mango in a given market area with the help of a dataset
provided by the government. The high accuracy rate achieved
during testing in stills confidence in the model’s reliability
and efficacy, underscoring its potential as a valuable tool for
automated mango type differentiation. These promising
results lay a solid foundation for further research and
development aimed at extending the model’s capabilities to Fig.5
encompass a broader range of mango varieties and enhance its
applicability in real-world scenarios. VI. FUTURE SCOPE

1. Expansion and Accuracy: Enhancing the dataset with more

mango species and improving model accuracy for better
classification under varying environmental conditions.

2. Additional Features: Integrating ripeness detection and

disease identification to aid in agricultural practices, providing
comprehensive support for mango quality management.

3.Real-Time Processing & Scalability: Developing mobile

applications and cloud-based solutions for real-time mango
classification, enabling wider accessibility and scalability for
farmers, suppliers, and traders.
Fig.2

VII. CONCLUSION
In conclusion, the proposed methodology for image-based
mango type differentiation has been significantly augmented
with the incorporation of price prediction and quality grading
capabilities. By seamlessly integrating convolutional neural
networks (CNNs) and advanced data pre-processing
techniques, our system not only achieves precise mango
variety classification but also offers invaluable insights into
market pricing dynamics and intrinsic fruit quality. This
comprehensive approach addresses the longstanding
challenges of manual classification and pricing uncertainties
in the mango industry. The addition of price prediction and
Fig.3 quality
grading features represents a pivotal advancement in mango
classification technology, promising to revolutionize industry
standards and significantly enhance quality control measures.
By fostering streamlined supply chain management and
elevating customer satisfaction levels, our innovative solution
catalyzes tangible improvements in the agroindustry
landscape, paving the way for sustainable
growth and prosperity.
VIII. REFERENCES
[1] Krithi D M Kshama Kalkur Megha A. Girish S, Jnana
Shetty K. Mango classification for agro-based industries using
transfer learning technique. volume 08, 2021.
Fig.4
[2] Prof. J. Girija Rohan Sriram, Amar Tejas M. Mango
classification using convolutional neural networks. volume 05,
2020.
[3] Bipan Koley Chiranjib. Nandi, Chandra Tudu. A machine
vision-based maturity prediction system for sorting of
harvested mangoes. IEEE Transactions on Instrumentation
and Measurement., (63):1722–1730, 2019.
[4] Duc Thong N. Cong H. T. Thanh Phong N. T. Thinh, N. T.
Mango classification system based on machine vision and
artificial intelligence. 7th International Conference on Control,
Mechatronics and Automation (ICCMA),2019.
[5] Duc Nguyen T. Dang N. T. Vo, M. T. Embedded machine
learning for mango classification using image processing and
support vector machine. 6th NAFOSTED Conference on
Information and Computer Science (NICS).,2019.
[6] Ashtekar S. Jayakrishna S. S. Bharath K. P. Vanathi P. T.
Rajesh Kumar M. Kumar, P. Classification of mango leaves
infected by fungal disease 42 MangoSpecs using deep
learning. 5th International Conference on Computing
Methodologies and Communication (ICCMC), 2021.
[7] Arora S. Kour, V. P. Particle swarm optimization based
support vector machine (p-svm) for the segmentation and
classification of plants. 2019.
[8] Islam saiful Zarrin, I. Leaf based trees identification using
convolutional neural network. IEEE 5th International
Conference for Convergence in Technology (I2CT), 2019.
[9] Mogensen L. V. Galeazzi R. Andersen J. C. Christensen, J.
H. Detection,localization and classification of fish and fish
species in poor conditions using convolutional neural
networks. 2018.
[10] Basu A. Nandy A. Deb A. Chowdhury R. Chanda D
Biswas, R. Identification of pathological disease in plants
using deep neural networks - powered by intel® distribution
of openvino™ toolkit. Indo – Taiwan 2nd International
Conference on Computing, Analytics and Networks (Indo
Taiwan CAN), 2020.
[11] Hsiao-Yen Tung Shih-Lun Wu and Yu-Lun Hsu. Deep
learning for automatic quality grading of mangoes: Methods
and insights. Indo – Taiwan 2nd International Conference on
Computing, Analytics and Networks (Indo-TaiwanICAN),
2020.
[12] Abeer A. Elsharif Samy S. Abu-Naser Alaa Soliman Abu
Mettleq, Ibtesam M. Dheir. Mango classification using deep
learning. International Journal ofAcademic Engineering
Research (IJAER), 2019.
[13] Ayesha Hakim Hafiz Muhammad Rizwan Iqbal.
Classification and grading of harvested mangoes using
convolutional neural network. International Journalof Fruit
Science, 2022.43MangoSpecs
[14] Wakode M. Supekar, A. Ripeness, size and shape based
automated mango grading using image processing and
machine learning techniques. International Journal of Current
Engineering and Technology, 2021.
[15] Wakode M. Supekar, A. Computer vision based
automated mango grading a review. Journal of Postharvest
Technology, 2020.
[16] Le-Thi D.-N. Doan, T.-N. A novel mango grading system
based on image processing and machine learning methods.
International Journal of Advanced Computer Science and
Applications, 2023.