datasets =https://www.kaggle.com/datasets/muratkokludataset/rice-image-dataset
Rice Grain Classification using MobileNetV2 This project implements a CNN-based rice grain classification model using MobileNetV2 to differentiate between five rice varieties: Arborio, Basmati, Ipsala, Jasmine, and Karacadag. The dataset consists of 75,000 images (15,000 per class). Dataset Preparation The image dataset is divided into training (80%) and validation (20%) sets using ImageDataGenerator.
Training Set: 80% of the images, with data augmentation applied (rotation, zoom, shear, and horizontal flip) to improve model generalization.
Validation Set: 20% of the images, only rescaled for evaluation.
🚀 Project Overview Model Used: MobileNetV2 (Pretrained on ImageNet)
Dataset: 75,000 rice grain images
Input Size: 224x224x3
Classes: 5
Evaluation Metrics: Accuracy, Precision, Recall, F1-Score
Hardware Used: CPU (Training on GPU recommended for faster processing)
📊 Performance Metrics Class Precision Recall F1-Score Support Arborio 1.00 0.96 0.98 3000 Basmati 0.99 0.99 0.99 3000 Ipsala 0.99 1.00 1.00 3000 Jasmine 0.99 0.99 0.99 3000 Karacadag 0.97 1.00 0.98 3000 Overall Accuracy 99% 99% 99% 15000 ⏳ Training Details Epochs: 2
Training Time:
Epoch 1: ~55 min (Accuracy: 89.58%, Val Acc: 98.62%)
Epoch 2: ~41 min (Accuracy: 96.95%, Val Acc: 98.71%)
Loss: Gradually decreased, indicating model convergence
⚡ Optimization Suggestions 🔹 Use a GPU (e.g., NVIDIA CUDA) to significantly reduce training time 🔹 Reduce image resolution (e.g., 128x128) for efficiency 🔹 Data Augmentation can further improve model robustness
Observations:
The model performs very well, with most predictions on the diagonal (correct classifications).
Misclassifications:
Arborio: 99 samples were misclassified.
Jasmine: Has some confusion with Basmati (22 samples misclassified).
Basmati & Ipsala: Almost perfect classification. Future Improvements 📌 Data Augmentation: Further augmentation techniques (rotation, scaling, flipping) can help improve robustness. 📌 Hyperparameter Tuning: Fine-tuning MobileNetV2 architecture (learning rate, dropout) may enhance performance. 📌 Model Pruning & Quantization: Optimize for mobile and edge deployments without significant accuracy loss.
Future Improvements 📌 Data Augmentation: Further augmentation techniques (rotation, scaling, flipping) can help improve robustness. 📌 Hyperparameter Tuning: Fine-tuning MobileNetV2 architecture (learning rate, dropout) may enhance performance. 📌 Model Pruning & Quantization: Optimize for mobile and edge deployments without significant accuracy loss.