PomegradeAI — AI-Powered Pomegranate Quality Grading System | Final Year Project with Source Code

PomegradeAI — Pomegranate Quality Grading System

PomegradeAI is a production-ready AI final year project built for agricultural quality assessment. It combines deep learning, REST API development, and cross-platform mobile application development into a single full-stack solution. The system automatically classifies pomegranates into three standardized quality grades using a fine-tuned EfficientNetB0 convolutional neural network, achieving a test accuracy of 97.4 percent. Whether you are a BTech CSE student looking for an AI final year project with source code or an MCA student who wants a unique machine learning project, PomegradeAI covers every dimension of a modern, real-world AI system.

This project is listed on CodeAj Marketplace, where you get the complete source code, pre-built project report, and optional mentorship support until your final submission.

What PomegradeAI Does

A user opens the Flutter mobile application, captures or uploads an image of a pomegranate, and the app sends that image over HTTP to a Django REST API. The backend preprocesses the image, runs inference through the EfficientNetB0 model, and returns a quality grade, confidence score, and a detailed recommendation — all within seconds. Every result is stored in a local SQLite database and accessible through a grading history screen in the app.

Quality Grade Classification

The model classifies pomegranates into the following three grades, each with tailored agricultural recommendations:

• QA Class 1 — High Quality (Export Grade): Suitable for direct export, premium retail, and gift packaging. Target price range is 80 to 120 rupees per kilogram, with a shelf life of 15 to 20 days when stored at 5 to 8 degrees Celsius.
• QB Class 2 — Medium Quality (Processing Grade): Ideal for juice extraction, anardana (dried seeds), and local markets. Storage at 8 to 12 degrees Celsius gives a shelf life of 7 to 10 days.
• QC Class 3 — Low Quality (Utilization Grade): Directed toward peel extraction for tannins and dyes, animal feed, biogas, or composting. Should be processed within 2 to 3 days.

Deep Learning Model Architecture

The model is built on EfficientNetB0 pre-trained on ImageNet, with a custom classification head added on top. The architecture flows from a 224x224x3 input through the EfficientNetB0 backbone, followed by GlobalAveragePooling2D, BatchNormalization, a Dense layer of 256 units with ReLU activation and 0.5 dropout, and finally a Softmax output layer with three nodes corresponding to the three quality grades.

Training was done in two phases. Phase 1 froze the backbone and trained only the classification head for 14 epochs at a learning rate of 1e-4. Phase 2 unfroze the top backbone layers and fine-tuned for 19 additional epochs at 1e-5. Both phases used Adam optimizer, categorical cross-entropy loss, EarlyStopping with patience of 5, ReduceLROnPlateau, and ModelCheckpoint callbacks.

The dataset contains 769 images across three classes — 391 high-quality, 253 medium-quality, and 125 low-quality. Class imbalance was handled using balanced sample weights. Augmentation techniques including random rotations, zoom, brightness and contrast shifts, and horizontal and vertical flips were applied through a tf.data pipeline.

Project Features

• 97.4 percent test accuracy on a three-class pomegranate quality dataset
• Two-phase transfer learning strategy with EfficientNetB0
• Django REST API with endpoints for grading, history, stats, and health check
• Flutter mobile app compatible with Android and iOS
• Grade-specific agricultural recommendations with price ranges and storage guidance
• Grading history stored in SQLite database
• Aggregate statistics endpoint showing grade distribution and average confidence
• Test-time augmentation (TTA) support for improved robustness
• Grad-CAM visualization in the training notebook for model explainability
• Modular codebase with separate preprocessor, inference, and recommendation layers

Real-World Applications

PomegradeAI demonstrates how AI can transform traditional agricultural supply chains. The system has direct applications in:

• Export-quality fruit sorting and grading at packing houses
• Cold storage management with grade-based shelf life recommendations
• Reducing post-harvest losses by routing low-grade produce to value-added processing
• Integration into smart farming platforms and agri-tech mobile applications
• Academic and research purposes in computer vision applied to agriculture

System Architecture Overview

The project follows a client-server architecture. The Flutter mobile app acts as the client, communicating over HTTP with the Django REST backend. The backend handles image preprocessing using OpenCV and Pillow, runs the TensorFlow/Keras inference pipeline, queries and writes results to SQLite, and returns structured JSON responses to the app. The entire stack is containerizable and can be deployed on a Linux VPS with Gunicorn serving the Django application behind Nginx.

API Reference Summary

• POST /api/grade/ — Submit a pomegranate image for grading. Returns grade label, confidence, recommendation, and all class probabilities.
• GET /api/history/ — Retrieve the last 20 grading results.
• GET /api/stats/ — Get aggregate counts, average confidence, and grade distribution.
• GET /health/ — Backend health check endpoint.

Who Should Buy This Project

This project is best suited for BTech CSE, MCA, and BSc IT students who want a Python AI final year project with source code that stands out in viva and presentations. It covers computer vision, transfer learning, REST API development, and mobile app development — giving you a project that spans multiple domains and demonstrates real-world relevance.

You can explore more projects like this — including other machine learning final year projects and Flutter-based final year projects — on the CodeAj Marketplace. If you need a project on a different domain or a fully custom solution, check out our custom project development service.

Add-On Services Available

• Idea Implementation: Have a unique project idea? The CodeAj team will build it from scratch for you.
• Project Setup and Source Code Explanation: A one-on-one session to set up the project on your machine and walk you through every module of the code so you can explain it confidently in your viva.
• Custom Project Report, Research Paper, and PPT: Get a college-format project report, a publishable research paper, and a professional presentation — all tailored to your project and your institution's requirements.