The title slide introduces the "Automated Rice Grain Classification System Using Deep Learning" for varieties like IRRI-6 White, Super White, and 1509 Steam, as part of a Mid-Year Evaluation. It lists four group members (Abul Hassan, Syed Muhammad Alman Raza, Muhammad Muneeb Zafar, Muhammad Sarim Khan), supervisor Ms. Fakhra Aftab, co-supervisor Dr. Majida Kazmi, and industrial partner MATCO Foods.

Rice Varieties: IRRI-6 White | Super White | 1509 Steam

Group Members: Abul Hassan (CS-22118) — CGPA: Syed Muhammad Alman Raza (CS-22127) — CGPA: Muhammad Muneeb Zafar (CS-22129) — CGPA: Muhammad Sarim Khan (CS-22142) — CGPA:

Supervisor: Ms. Fakhra Aftab Co-Supervisor: Dr. Majida Kazmi Industrial Partner: MATCO Foods

Mid-Year Evaluation

Source: Final Year Project Mid-Year Evaluation

This slide emphasizes the importance of rice variety classification in industry and the limitations of manual inspection methods. It outlines the role of computer vision and deep learning in automating the process for greater accuracy and efficiency.

Project Background & Motivation

• Importance of rice variety classification in industry
• Limitations of manual inspection methods
• Role of computer vision and deep learning
• Objective: Automate for improved accuracy and efficiency

The slide highlights three target rice varieties—IRRI-6 White, Super White, and 1509 Steam—noted for their fine grains, premium milling quality, aroma, and industrial demand. It also underscores their essential role in processing/export quality control and the challenge of manual differentiation due to subtle visual similarities.

Target Rice Varieties

{ "features": [ { "icon": "🌾", "heading": "IRRI-6 White", "description": "Fine-grained white rice variety with high industrial demand and consistent quality." }, { "icon": "🌾", "heading": "Super White", "description": "Premium white rice prized for its superior milling quality and market value." }, { "icon": "🌾", "heading": "1509 Steam", "description": "Long-grain Basmati variety ideal for steaming, valued for aroma and texture." }, { "icon": "🏭", "heading": "Industrial Relevance", "description": "Essential for quality control in rice processing and export industries." }, { "icon": "👁️", "heading": "Visual Similarity", "description": "Challenging manual differentiation due to subtle visual differences." } ] }

The slide presents a workflow for rice grain variety classification, starting with image capture in a Custom Black Chamber, instance segmentation via YOLOv8-Seg/YOLOv11-Seg, and CNN classification using VGG16/19, ResNet50, MobileNetV2, or InceptionV3 to identify IRRI-6 White, Super White, or 1509 Steam varieties. It concludes with real-time predictions delivered through an Android mobile app.

System Overview

{ "headers": [ "Stage", "Key Component", "Description" ], "rows": [ [ "Image Capture", "Custom Black Chamber", "Controlled lighting and fixed camera position ensures consistent image quality" ], [ "Instance Segmentation", "YOLOv8-Seg / YOLOv11-Seg", "Pixel-level separation of individual rice grains from captured images" ], [ "CNN-based Classification", "VGG16/19, ResNet50, MobileNetV2, InceptionV3", "Classifies segmented grains into IRRI-6 White, Super White, or 1509 Steam varieties" ], [ "Mobile App Prediction", "Android Application", "Provides real-time variety prediction on user-captured images" ] ] }

Source: Automated Rice Grain Classification System

The slide "Hardware Setup – Image Acquisition Chamber" shows a custom black chamber for image acquisition. It features a controlled lighting environment, fixed camera position, and ensures consistent image quality.

Hardware Setup – Image Acquisition Chamber

!Image

• Custom black chamber design
• Controlled lighting environment
• Fixed camera position
• Ensures consistent image quality

Source: Wikipedia

The slide outlines a custom dataset with 2000 images captured using a custom-designed chamber, covering IRRI-6 White, Super White, and 1509 Steam rice varieties. These images are prepared for instance segmentation and classification to ensure high-quality data for deep learning models.

Custom Dataset Development

• 2000 images captured using custom-designed chamber
• Covers IRRI-6 White, Super White, 1509 Steam
• Prepared for instance segmentation and classification
• Ensures high-quality data for deep learning models

Source: Final Year Project Presentation

This slide, titled "Instance Segmentation Using Roboflow," showcases an image of accurately separated rice grains with pixel-level annotations. It highlights Roboflow's features for data augmentation, dataset robustness, and efficient management.

Instance Segmentation Using Roboflow

!Image

• Roboflow enables pixel-level annotation of images
• Provides data augmentation for dataset robustness
• Facilitates efficient dataset management
• Separates individual rice grains accurately

Source: Wikipedia

The slide outlines training YOLOv8-Seg and YOLOv11-Seg models on a public rice grain dataset. It reports establishing baseline validation metrics with successful detection and segmentation.

Model Training – Public Dataset

• Trained YOLOv8-Seg and YOLOv11-Seg models
• Utilized public rice grain dataset
• Established baseline validation metrics
• Achieved successful detection and segmentation

YOLOv11 achieves 96.3% mAP@0.5 on a custom dataset, delivering superior segmentation accuracy over YOLOv8's strong 92.5% baseline. This marks a +18.4% improvement versus public benchmarks, validating the dataset's quality.

Model Training – Custom Dataset

• 96.3%: YOLOv11 mAP@0.5

Superior segmentation accuracy

• 92.5%: YOLOv8 mAP@0.5

Strong baseline performance

• +18.4%: Improvement vs Public

Validates custom dataset quality

The slide "Main Target – Classification Strategy" outlines a two-step pipeline in table format. It uses VGG16, VGG19, ResNet50, MobileNetV2, and InceptionV3 for classifying segmented grains, plus a hybrid CNN-SVM approach for enhanced accuracy.

Main Target – Classification Strategy

{ "headers": [ "Pipeline Step", "Models / Approach" ], "rows": [ [ "Segmented grains to classifier", "VGG16, VGG19, ResNet50, MobileNetV2, InceptionV3" ], [ "Hybrid for accuracy", "CNN-SVM" ] ] }

The hardware chamber is completed, a custom dataset of 2000 images has been created, and YOLO models have been trained on both public and custom datasets. The custom dataset yields higher accuracy, with the classification stage currently in progress.

Current Progress Summary

• Hardware chamber completed
• Custom dataset of 2000 images created
• YOLO models trained on public dataset
• YOLO models trained on custom dataset
• Custom dataset shows higher accuracy
• Classification stage in progress

The slide outlines future plans to expand the custom dataset, train and compare CNN classifiers (e.g., VGG, ResNet, MobileNet), develop an Android app, and conduct final testing with MATCO Foods. It concludes with strong mid-year progress, a validated segmentation pipeline, and the project on track for success, followed by thanks and questions.

Future Plan & Conclusion

**Future Plan:

• Expand custom dataset
• Train and compare CNN classifiers (VGG, ResNet, MobileNet, etc.)
• Develop Android mobile application
• Final testing with MATCO Foods

Conclusion:

• Strong mid-year progress achieved
• Segmentation pipeline validated
• Project on track for final completion

Thank you for your attention!

Questions?**

On Track for Success

Source: Final Year Project Mid-Year Evaluation